JP3549700B2 - Image processing device - Google Patents

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an image processing apparatus for compressing and converting multi-tone image data into multi-valued quantized image data.
[0002]
[Prior art]
2. Description of the Related Art Conventionally, there is an image processing apparatus that converts input multi-tone image data into binary image data using, for example, an error diffusion method. In this error diffusion method, a binarization quantization error between a gradation value of a target pixel and a fixed binarization threshold value for performing a binarization process on the gradation value is calculated by calculating a binarization error of a pixel near the target pixel. This is a method of sequentially diffusing each gradation value into binary values by diffusing the gradation values.
[0003]
A basic technique relating to error diffusion is described in, for example, “Proceeding of the S.A. I. D. vol 17/2, 1976, R.C. Floyd and L. Steinberg "An Adaptive Algorithm for Spatial Grayscale" pp. 75-76] and "Proceedings of the IEEE, vol. 76, No. 1, January 1988, Robert A. Urichney “Dithering with Blue Noise” pp. 56-79].
[0004]
By extending algorithms adapted from these documents and setting a fixed number of quantization thresholds and quantization levels according to the required number of quantizations, it is possible to realize multi-level quantization by multi-level error diffusion processing. it can.
[0005]
The multi-level quantization technique is also disclosed, for example, in Japanese Patent Application Laid-Open No. 3-192970. The image processing apparatus disclosed in the above publication includes a plurality of quantization means for quantizing input multi-tone image data into multi-values having different numbers of levels by pseudo halftone processing. One of the plurality of quantization means is selected based on the characteristic. This makes it possible to perform desired multi-value pseudo halftone processing on both the character portion and the halftone image.
[0006]
[Problems to be solved by the invention]
However, the configuration of a conventional image processing apparatus that performs multi-level quantization using the error diffusion method has the following problems.
[0007]
For example, in a gradation image in which gradation changes gradually, similar image densities (gradation values) are continuous. That is, the gradation image has image data in which the image density near the intermediate quantization level is continuous. Therefore, when such image data is quantized, as a characteristic of the basic algorithm, a change in texture (fine pattern pattern) becomes apparent due to the continuation of the same intermediate quantization level between adjacent pixels. I do. As a result, a pseudo contour different from the original contour is generated and may be recognized as the original contour.
[0008]
Further, in the gradation image, a stripe pattern which is generally called “Worm” and is characteristic of the multi-level error diffusion processing appears in a portion where the image density is high or a portion where the image density is low.
[0009]
SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and an object of the present invention is to provide a good quantization process without generating pseudo contours and "Worm" even in an image having a gradual gradation change. It is an object of the present invention to provide an image processing apparatus that can perform the processing.
[0010]
[Means for Solving the Problems]
According to a first aspect of the present invention, there is provided an image processing apparatus for converting input multi-tone image data into multi-valued quantized image data. And a feature value extracting means for extracting a feature value of the pixel of interest based on the pixel values of the pixels in the vicinity thereof and determining a region to which the pixel of interest belongs based on the feature value. A quantization threshold generation unit for generating a quantization threshold, a quantization level generation unit for generating a quantization level for performing a quantization process on the pixel of interest, and a comparison between a pixel value of the pixel of interest and the quantization threshold Thus, the quantization processing means for quantizing the pixel value of the target pixel to a predetermined quantization level, the diffusion range of the quantization error generated in the quantization processing by the quantization means, and the diffusion range And a quantization error diffusion means for performing an error diffusion process, wherein the quantization threshold value generation means and the quantization level generation means correspond to the region information of the target pixel. Generates the quantization threshold and the quantization levelFurther, when it is determined that the pixel of interest belongs to “photograph” or “indistinguishable area”, the position of the pixel of interest is set to (i , j), the quantization level generation means selects the combination of the quantization levels based on the remainder of (i + j) / 3.It is characterized by:
[0011]
According to the above configuration, the quantization threshold for quantizing the target pixel is generated by the quantization threshold generation unit. Further, a quantization level for performing a quantization process on the target pixel is generated by a quantization level generation unit. Then, the quantization processing means compares the quantization threshold value with the pixel value of the target pixel, and quantizes the pixel value of the target pixel to a predetermined quantization level. Then, the quantization error generated at the time of the quantization processing is diffused by the quantization error diffusion means.
[0012]
At this time, since the quantization level is generated in accordance with the region information of the pixel of interest generated by the feature amount extraction means, the same quantization level must be continuous between the pixel of interest and its neighboring pixels. Is almost gone. In addition, since the quantization threshold is also generated according to the region information of the pixel of interest generated by the feature amount extraction unit, the quantization threshold and the quantization threshold are used to generate the quantization threshold. Pixel quantization processing is adaptively performed in accordance with the target pixel.
[0013]
Therefore, according to the above configuration, even if the processed image is a gradation image having a gradual change in gradation, the generation of texture due to the continuation of the same quantization level and the multi-level error diffusion specific The occurrence of false contours can be reduced, and a good pseudo tone reproduction image can be obtained.
[0014]
According to another aspect of the present invention, there is provided an image processing apparatus for converting input multi-tone image data into multi-valued quantized image data.(I, j)And neighboring pixels(I + 1, j)A feature value extraction means for extracting a feature value of the pixel of interest based on the pixel value of the above, and determining a region to which the pixel of interest belongs based on the feature value; and a first quantization for quantizing the neighboring pixel. A threshold value generating means for generating a second quantization threshold value for performing a quantization process on the pixel of interest and a first quantization level for performing a quantization process on the neighboring pixels; A quantization level generating means for generating a second quantization level for performing a quantization process on the pixel of interest, and comparing the pixel value of the neighboring pixel with the first quantization threshold to obtain a pixel value of the neighboring pixel. Is quantized to a predetermined quantization level, and the pixel value of the pixel of interest is quantized to a predetermined quantization level by comparing the pixel value of the pixel of interest with the second quantization threshold. And the quantization hand Quantization error diffusion means for defining a diffusion range of a quantization error generated in the quantization process, determining a diffusion coefficient for each pixel within the diffusion range, and performing an error diffusion process. The processing means is (Pixel (i + 1, j) on line (i + 1)In addition to performing the first multi-level quantization process, the quantization error diffusion unit performs the first multi-level quantization only on pixels that have not been subjected to the quantization process with respect to the pixels on the line including the target pixel. Performing an error diffusion process for diffusing a first quantization error generated by the process, and thereafter, the quantization processing means includes:About the pixel (i, j) on the i-th lineIn addition to performing the second multi-level quantization processing, the quantization error diffusion unit performs the second multi-level quantization only on pixels that have not been subjected to the quantization processing with respect to the pixels on the line including the target pixel. It is characterized in that an error diffusion process for diffusing a second quantization error generated by the process is performed.
[0015]
According to the above configuration, the first quantization threshold for quantizing the neighboring pixels and the second quantization threshold for quantizing the pixel of interest are generated by the quantization threshold generation unit. Further, a first quantization level for quantizing the neighboring pixels and a second quantization level for quantizing the pixel of interest are generated by the quantization level generating means.
[0016]
Here, the quantization processing unit and the quantization error diffusion unit perform the quantization process and the error diffusion process on the line including the neighboring pixel, and then perform the quantization process and the error diffusion process on the line including the target pixel. Do each. That is, the quantization processing means quantizes the pixel value of the neighboring pixel to a predetermined quantization level by comparing the first quantization threshold with the pixel value of the neighboring pixel, and the quantization error diffusion means The quantization error generated at the time of the quantization process is diffused to neighboring pixels. Thereafter, the quantization processing means quantizes the pixel value of the pixel of interest to a predetermined quantization level by comparing the second quantization threshold with the pixel value of the pixel of interest, and the quantization error diffusion means The quantization error generated at the time of the quantization process is diffused to neighboring pixels.
[0017]
By such processing, the diffusion form of the quantization error is easily disturbed. Therefore, even if the processed image is a gradation image having a gradual change in gradation, it is possible to avoid the occurrence of a stripe pattern called “Worm” unique to the multi-level error diffusion processing in a portion where the image density is high or a portion where the image density is low. And a good pseudo-tone reproduction image can be obtained.
[0018]
According to a third aspect of the present invention, in the image processing apparatus according to the second aspect, the quantization processing unit and the quantization error diffusion unit are configured to perform the quantization processing and the quantization error diffusion in a serpentine system. It is characterized in that each of the error diffusion processes is performed.
[0019]
According to the above configuration, the quantization process and the error diffusion process are not always performed in the same direction, but are performed in the opposite direction by inverting each line. Thus, it is possible to easily disturb the uniform diffusion direction of each quantization error, and it is possible to prevent "Worm" which is a defect of the error diffusion processing from occurring.
[0020]
According to a fourth aspect of the present invention, in the image processing apparatus according to the second or third aspect, the quantization threshold value generating means and the quantization level generating means are arranged so that the target pixel belongs to the image processing apparatus. The method is characterized in that the second quantization threshold and the second quantization level are generated according to the area information.
[0021]
According to the above configuration, since the second quantization level is generated in accordance with the belonging area information of the target pixel, the same quantization level is hardly continued in the target pixel and its neighboring pixels. In addition, since the second quantization threshold is also generated according to the region information of the pixel of interest, the quantization of the pixel of interest can be performed by using the second quantization level and the second quantization threshold. The processing is adaptively performed according to the target pixel.
[0022]
Therefore, according to the above configuration, even if the processed image is a gradation image having a gradual change in gradation, the generation of texture due to the continuation of the same quantization level and the multi-level error diffusion specific The occurrence of false contours can be reduced, and a good pseudo tone reproduction image can be obtained.
[0023]
According to a fifth aspect of the present invention, in the image processing apparatus according to any one of the first to fourth aspects, the quantization level generating means is configured to perform the quantization processing near the pixel of interest. It is characterized in that a quantization level of a target pixel to be quantized is generated based on a quantization level of a pixel.
[0024]
According to the above configuration, the probability that the quantization level of the target pixel and the quantization level of the already-quantized pixel are continuously the same is reliably reduced. Therefore, even if the pixel of interest belongs to, for example, a "photograph" in which pseudo contours are particularly likely to occur, the probability of occurrence of pseudo contours can be reliably reduced, and a good quantized image can be reliably obtained.
[0025]
According to a sixth aspect of the present invention, in the image processing apparatus according to the fifth aspect of the present invention, in the image processing apparatus according to the fifth aspect of the present invention, the quantization level generating means generates a predicted quantization level based on a pixel value of the target pixel. After generating,When the predicted quantization level becomes the same as the intermediate quantization level,It is characterized in that the quantization level is changed to a quantization level different from the predicted quantization level based on the magnitude relationship between the pixel value and the set quantization level.
[0026]
According to the above configuration, since a quantization level different from the predicted quantization level predicted based on the pixel value of the target pixel is generated, the quantization level of the target pixel and its neighboring pixels is continuously the same. Probability is significantly reduced. As a result, the quantization error generated by the quantization process is also reduced. Therefore, even if the pixel of interest belongs to, for example, a "photograph" in which pseudo contours are particularly likely to occur, the probability of occurrence of pseudo contours can be greatly reduced, and a better quantized image can be reliably obtained.
[0027]
According to a seventh aspect of the present invention, in the image processing apparatus according to any one of the first to sixth aspects, the feature amount extracting means uses a neural network to belong to the target pixel. It is characterized in that the region is determined.
[0028]
According to the above configuration, by using the neural network, the region information of the target pixel can be accurately and accurately captured. Therefore, the quantization process adapted to each pixel can be performed with high accuracy and reliably.
[0029]
BEST MODE FOR CARRYING OUT THE INVENTION
[Embodiment 1]
One embodiment of the present invention will be described below with reference to FIGS.
[0030]
As shown in FIG. 1, the image processing apparatus according to the present embodiment includes a quantization unit 1 that performs a quantization process on a multi-tone input image obtained by a scanner (not shown) or the like.
[0031]
The quantization means 1 generally performs quantization processing of a multi-tone image by using either a fixed threshold or a variable threshold. When a fixed threshold value is used, it is general to set the threshold value to approximately the center value of the gradation values of the multi-tone image, but it is also possible to appropriately set the threshold value to another value according to the image to be processed. is there. On the other hand, when a fluctuation threshold is used, it is possible to appropriately disturb the threshold using a random threshold, a periodic or aperiodic function, or the like.
[0032]
In the present embodiment, the case where a fixed threshold is used in quaternary quantization will be described. However, the same algorithm as in the present embodiment is also used in the case where multi-level quantization of four or more values and a variable threshold are used. Thus, a quantization process can be performed. For convenience of explanation, it is assumed that the input pixel value is represented by 256 gradation values from 0 to 255, and that 0 represents black and 255 represents white.
[0033]
The quantization means 1 includes a feature amount extraction means 2, a quantization threshold value generation means 3, a quantization level generation means 4, a quantization processing means 5, a quantization error recognition means 6, and a quantization error diffusion means 7. .
[0034]
The feature amount extraction unit 2 is configured to calculate a feature such as a sum or a difference of these pixel values based on a pixel to be processed (hereinafter, referred to as a pixel of interest) in an image having a continuous tone value and a pixel value of a neighboring pixel. It is a means to determine the quantity. Then, based on the correlation between the feature amounts, the feature amount extracting unit 2 determines whether the pixel of interest is “text (character)”, “text base”, “photograph”, “halftone dot”, and “indistinguishable area”. The determination result is output to the quantization threshold generation unit 3 and the quantization level generation unit 4 as belonging region information.
[0035]
Note that the feature amount is not limited to the sum or difference of pixel values, but may be any value that can be obtained mathematically.
[0036]
The quantization threshold generation unit 3 is a unit that sets a quantization threshold for performing a quantization process on the target pixel according to the belonging area information of the target pixel output from the feature amount extraction unit 2. That is, when the feature amount extraction unit 2 determines that the target pixel belongs to, for example, “text” or “text base”, the quantization threshold generation unit 3 is set by the quantization level generation unit 4 as described later. The quantization threshold value is set to approximately the median value (128) of the upper and lower two quantization levels (0 and 255). On the other hand, when the feature amount extraction unit 2 determines that the pixel of interest belongs to, for example, a “dot”, a “photo”, or an “indistinguishable area”, the quantization threshold generation unit 3 outputs three quantization thresholds ( For example, 43, 128, and 213) are set.
[0037]
The quantization processing unit 5 performs multi-level quantization by comparing the pixel value of the target pixel with the quantization threshold set as described above, and outputs a multi-level quantized image. The quantization error recognizing means 6 recognizes a quantization error generated by the quantization processing in the quantization processing means 5 and outputs the result to the quantization error diffusion means 7.
[0038]
The quantization error diffusion means 7 defines a diffusion range of the quantization error recognized by the quantization error recognition means 6, determines a diffusion coefficient indicating an error weight for each pixel in the diffusion range, and performs error diffusion. Perform processing. Then, in order to adaptively perform the quantization process for each pixel of the input image, the quantization error diffusion unit 7 outputs a signal based on the end of the error diffusion process to the feature extraction unit 2 after the end of the error diffusion process in a predetermined pixel. This is transmitted to the quantization threshold generation means 3. As a result, the quantization process is sequentially performed on the scheduled area of the original image.
[0039]
The quantization level generation unit 4 quantizes the target pixel according to the region information of the target pixel output from the feature amount extraction unit 2 so that the same quantization level does not continue in the target pixel and its neighboring pixels. This is a means for setting a quantization level for processing.
[0040]
Here, the four quantization levels are referred to as level 0, level 1, level 2, and level 3, respectively. When the feature amount extraction unit 2 determines that the target pixel belongs to, for example, “text” or “text base”, the quantization level generation unit 4 sets the level 0 = level 1 = 0 and the level 2 = level 3 = 255. Set to. When it is determined that the pixel of interest belongs to, for example, a “dot”, the quantization level generation unit 4 outputs the level 0 = 0, the level 1 = the level 2 = 85, the level 3 = 255, or the level 0 = 0. , Level 1 = level 2 = 170 and level 3 = 255. When it is determined that the pixel of interest belongs to, for example, a “photograph” or an “indistinguishable area”, the quantization level generation unit 4 determines whether the pixel of interest belongs to “halftone” by a method described later. , Level 0 = 0, level 1 = 85, level 2 = 170, and level 3 = 255. If it is determined that the pixel of interest does not belong to any of the above, the quantization level generation means 4 sets level 0 = 0, level 1 = 85, level 2 = 170, and level 3 = 255.
[0041]
When it is determined that the target pixel belongs to, for example, a “photograph” or an “indistinguishable area”, the quantization level generation unit 4 determines one of the above three options as follows. select.
[0042]
Option 1 when each level is a combination of (0, 85, 255), Option 2 when each level is a combination of (0, 170, 255), and Option 0 when each level is (0, 85, 170, 255) Are set as option 3, and the position of the target pixel is set as (i, j). Here, when the remainder of (i + j) / 3 is 0, the quantization level generation means 4 selects option 1. On the other hand, when the remainder of (i + j) / 3 is 1, the quantization level generation means 4 selects option 2. When the remainder of (i + j) / 3 is 2, the quantization level generation unit 4 selects option 3.
[0043]
When the pixel of interest belongs to, for example, a "halftone dot", the quantization level generation means 4 sets an appropriate quantization level based on, for example, the remainder of (i + j) / 2 in the same manner as described above.
[0044]
Next, the quantization processing in the image processing apparatus according to the present embodiment will be specifically described.
[0045]
First, when an image to be processed is input, the target pixel is set to “text” or “text base” by the characteristic amount extraction means 2 based on the characteristic amount obtained from the pixel value of the target pixel and its neighboring pixels. , "Photograph", "halftone dot", and "indistinguishable area". Then, the quantization threshold is set by the quantization threshold generator 3 as described above based on the determination result by the feature amount extractor 2.
[0046]
Subsequently, the quantization level is set by the quantization level generation unit 4 as described above based on the belonging area information from the feature amount extraction unit 2, and the quantization processing unit 5 performs the quantization process on the target pixel. Done.
[0047]
The quantization error generated during the quantization processing by the quantization processing means 5 is recognized by the quantization error recognition means 6 and sent to the quantization error diffusion means 7. The quantization error diffusion means 7 determines a diffusion range and a diffusion coefficient of the quantization error and performs an error diffusion process, for example, as shown in FIG.
[0048]
In FIG. 2, the main scanning direction is indicated by an arrow j, and the sub scanning direction is indicated by an arrow i. Assuming that the position of the pixel of interest is (i, j), the diffusion range determined by the quantization error diffusion means 7 is pixel (i, j + 1), (i + 1, j-1), (i + 1, j), (I + 1, j + 1), and the respective diffusion coefficients are 、, ８, ４, １ ／.
[0049]
When the error diffusion processing to the neighboring pixels is completed by the quantization error diffusion means 7, the above processing is repeated as necessary until the quantization processing of the planned area of the original image is completed.
[0050]
According to the above configuration, parameters such as the quantization threshold value and the quantization level of the target pixel are set according to the region information of the target pixel, and the quantization process is performed. Can be performed adaptively.
[0051]
Moreover, since the quantization level of the target pixel is set by the quantization level generation means 4 so that the same quantization level does not continue in the target pixel and its neighboring pixels, the quantization levels are continuous between adjacent pixels. The occurrence of texture caused by the above is suppressed. Therefore, even if the processed image is a gradation image having a gradual change in gradation, the same quantization level is not continuous in adjacent pixels, so that the occurrence frequency of a pseudo contour peculiar to multi-level error diffusion is reduced, and Quantization processing can be performed.
[0052]
The quantization threshold, the quantization level, the combination thereof, and the like described in the present embodiment are not particularly limited, and can be changed according to various conditions such as the characteristics of the output device.
[0053]
[Embodiment 2]
Another embodiment of the present invention will be described below with reference to FIG.
[0054]
In the present embodiment, in the image processing apparatus described in the first embodiment, the quantization level generation unit 4 sets the quantization level of the pixel of interest to be quantized when the belonging area obtained from the feature amount extraction unit 2 is set. In addition to the information, the quantization level is generated based on the quantization level of the pixel quantized immediately before the target pixel, that is, the quantization level of the quantized pixel near the target pixel. I have. In the following, for simplicity of explanation, it is assumed that the feature pixel extraction unit 2 has determined that the pixel of interest belongs to, for example, “photograph”.
[0055]
If the quantization level of the quantized pixel in the vicinity of the pixel of interest is, for example, 0, the quantization level generation unit 4 performs the above-described processing based on the remainder of (i + j) / 3, as in the first embodiment. One of options 1 to 3 is selected. In other words, the quantization level generation means 4 selects one of the combinations in which each level consists of a combination of (0, 85, 255), a combination of (0, 170, 255), and (0, 85, 170, 255). Will do.
[0056]
On the other hand, when the quantization level of the already-quantized pixel is, for example, 85, the quantization level generation unit 4 selects the above-described option 2 (each level is a combination of (0, 170, 255)). Has become. When the quantization level of the already-quantized pixel is, for example, 170, the quantization level generation unit 4 selects the above-described option 1 (combination of each level of (0, 85, 255)). Has become. Further, when the quantization level of the already-quantized pixel is, for example, 255, the quantization level generation unit 4 performs the above-described option 1 based on the remainder of (i + j) / 3, as in the first embodiment. To 3 are selected.
[0057]
As described above, the quantization level generation unit 4 generates the quantization level for the target pixel in consideration of the setting status of the quantization levels of the quantized pixels around the target pixel. And the probability that the quantization level of the already-quantized pixel is continuously the same is reliably reduced. Therefore, even if the pixel of interest belongs to a “photograph” in which pseudo contours are particularly likely to occur, the probability of occurrence of pseudo contours can be reliably reduced, and a good quantized image can be reliably obtained.
[0058]
[Embodiment 3]
Another embodiment of the present invention will be described below with reference to FIG.
[0059]
In the present embodiment, in the image processing apparatus used in the second embodiment, after the quantization level generation unit 4 further generates a predicted quantization level based on the pixel value (image density, gradation value) of the pixel of interest. The quantization level is changed to a quantization level different from the predicted quantization level based on the magnitude relationship between the pixel value and the set quantization level.
[0060]
That is, the quantization level generation unit 4 predicts the quantization level based on the image density information of the pixel of interest, and then determines the magnitude relationship between the image density and the quantization level, and determines that the image density is greater than the quantization level. If determined, the desired quantization level is semi-forcibly changed to a quantization level larger than the predicted quantization level, while if the image density is determined to be smaller than the quantization level, the desired quantization level is set to the predicted quantization level. The quantization level is semi-forcibly changed to a quantization level smaller than the level.
[0061]
Specifically, it is assumed that the quantization level is currently set to (0, 85, 170, 255) and the quantization threshold is set to (43, 128, 213). When the image density of the pixel of interest is 190, the prediction quantization level is predicted to be 170 by the quantization level generation means 4. Here, usually, the quantization level of the target pixel to be quantized is set to 170 by comparing the image density 190 with the quantization threshold 213. However, in the present embodiment, since the image density 190 is higher than the quantization level 170, the quantization level generator 4 reduces the required quantization level to a quantization level (for example, 255) larger than the prediction quantization level 170. It is forcibly changed.
[0062]
According to the above configuration, when the predicted quantization level predicted based on the pixel value of the pixel of interest is the same as the quantization level obtained based on the magnitude relationship between the pixel value and the quantization threshold, The quantization level generating means 4 generates a quantization level different from the predicted quantization level. As a result, the probability that the quantization level of the target pixel and its neighboring pixels will be continuously the same will be significantly smaller than in the case of the second embodiment.. AndTherefore, even if the pixel of interest belongs to a “photograph” in which pseudo contours are particularly likely to occur, the probability of occurrence of pseudo contours is greatly reduced, and a better quantized image can be reliably obtained.
[0063]
Note that the processing of this embodiment is performed only when the predicted quantization level is the same as the two intermediate quantization levels (here, 85 and 170).
[0064]
[Embodiment 4]
Another embodiment of the present invention will be described below with reference to FIGS.
[0065]
In the present embodiment, in the image processing apparatus described in the first to third embodiments, the feature amount extracting unit 2 determines “density difference”, “complexity”, “run” based on image density information of a target pixel and its neighboring pixels. Information such as "length" is obtained, and a feature amount is extracted from the correlation using a neural network. Then, the quantization threshold generation unit 3 and the quantization level generation unit 4 set the quantization threshold and the quantization level, respectively, based on the feature amount.
[0066]
As shown in FIG. 3, the neural network has three inputs and three outputs, and is composed of an input layer 11, an intermediate layer 12, and an output layer 13. The input layer 11 includes three nodes: a node A1 to which the input value X1 is input, a node A2 to which the input value X2 is input, and a node A3 to which the input value X3 is input. The intermediate layer 12 includes six nodes B1 to B6, each of which is coupled to each of the nodes A1 to A3 of the input layer 11, respectively. The output layer 13 includes three nodes C1 to C3, each of which is coupled to each of the nodes B1 to B6 of the intermediate layer 12, respectively.
[0067]
The number of nodes in the intermediate layer 12 is calculated by increasing the number of nodes in the intermediate layer 12 one by one in order from 1 using the sample image data. To be used. Therefore, as in the present embodiment, the number of nodes in the intermediate layer 12 is not limited to six, and may be determined in consideration of learning accuracy.
[0068]
The input values X1, X2, and X3 correspond to information of "density difference", "complexity", and "run length" obtained based on image density information of the target pixel and its neighboring pixels. When the values X1, X2, and X3 are input to the input layer 11, analog output values Y1, Y2, and Y3 are output from the nodes C1 to C3 of the output layer 13 via the intermediate layer 12. I have.
[0069]
Here, in the present embodiment, for example, when the information that the “density difference” is large, the “complexity” is small, and the “run length” is large is given to the input layer 11 as the input values X1, X2, and X3, respectively. By the operation from the input layer 11 to the output layer 13, the output corresponding to the "text" portion is activated as compared with the other portions. For example, when information indicating that “density difference” and “complexity” are relatively large and “run length” is relatively small are given to the input layer 11 as input values X1, X2, and X3, respectively, The output corresponding to the "halftone dot" portion is activated by the calculation from the output layer 13 to the output layer 13 as compared with the other portions. Further, for example, when information that “density difference” and “complexity” are relatively small and “run length” is relatively large are given to the input layer 11 as input values X1, X2, and X3, respectively, By the calculation from to, the output corresponding to the “photograph” portion is activated as compared with the other portions.
[0070]
As described above, by using the neural network, the pixel of interest as to which of “text”, “text background”, “photograph”, “halftone dot”, and “indistinguishable area” belong can be determined. The belonging area information can be accurately and accurately captured. Therefore, the quantization process adapted to each pixel can be performed with high accuracy and reliably.
[0071]
In this embodiment, a plurality of parameters necessary for the error diffusion processing are prepared. The parameters are the quantization threshold, the quantization level, the diffusion range of the quantization error, the diffusion coefficient, and the reference range of the quantization level as shown in FIGS. 2, 4, and 5. Each parameter is adaptively selected according to the region information of the pixel of interest determined by the feature amount extraction means 2, and quantization processing is performed.
[0072]
That is, when the feature amount extraction unit 2 determines that the pixel of interest belongs to, for example, “text”, the quantization error diffusion unit 7 sets the diffusion range and diffusion coefficient of the quantization error as shown in FIG. Selected. On the other hand, when it is determined that the pixel of interest belongs to, for example, “photograph”, the quantization level generating means 4 similarly selects a diffusion range and a diffusion coefficient of the quantization error as shown in FIG. As a result, it is possible to perform the quantization process adapted to the region to which the target pixel belongs without increasing the process amount and the required memory amount of the entire processing system.
[0073]
Further, the above-described neural network learns based on inputting image data formed using, for example, characters, photographs, halftone dots, or the like as teacher data into the input layer 11 in advance. . As a result, the accuracy of determining the image region containing the pixel of interest is improved, and the output ratio of the “non-discriminable region” is reduced. As a result, it is possible to improve the accuracy of the optimal quantization processing for the target pixel.
[0074]
When the input multi-tone image data has a specific pattern, specifically, when the absolute value of the image density of the target pixel is in a range of about 95% or more or about 5% or less of the maximum density value The feature amount extracting means 2 takes measures such as forcibly setting the quantization error to zero, and prevents the error diffusion processing by the quantization error diffusion means 7. This is one of the adverse effects of the error diffusion processing when the target pixel belongs to, for example, “text base”, that is, when the original image has a region close to solid white or solid black over a relatively wide range. In addition, it is possible to prevent the occurrence of an inversion region caused by accumulation of quantization errors.
[0075]
[Embodiment 5]
Another embodiment of the present invention will be described below with reference to FIGS.
[0076]
In the present embodiment, in the image processing apparatus described in the first embodiment, the quantization threshold value generation unit 3 and the quantization level generation unit 4 quantize the neighboring pixel (i + 1, j) of the pixel of interest (i, j). A first quantization threshold and a first quantization level for processing are adaptively selected according to the belonging area information of the neighboring pixel (i + 1, j), and after the quantization processing of the neighboring pixel (i + 1, j). A second quantization threshold value and a second quantization level for quantizing the target pixel (i, j) are adaptively selected according to the region information of the target pixel (i, j). .
[0077]
The quantization processing means 5 compares the pixel value of the neighboring pixel (i + 1, j) with the first quantization threshold to perform the first multi-level quantization processing on the line of the neighboring pixel (i + 1, j). After the first multi-level quantization process, the pixel value of the pixel of interest (i, j) is compared with the second quantization threshold to determine the second line of the pixel of interest (i, j). Multi-level quantization processing is performed.
[0078]
The quantization error recognizing means 6 includes a first quantization error E1 (see FIG. 7A) generated by the first multi-level quantization processing and a second quantization generated by the second multi-level quantization processing. Recognize the errors E2 (see FIG. 7 (b)) and output them to the quantization error diffusion means 7.
[0079]
The quantization error diffusion means 7 defines the diffusion range of the first and second quantization errors E1 and E2 recognized by the quantization error recognition means 6, and indicates the weight of the error for each pixel within the diffusion range. The diffusion coefficient is determined, and an error diffusion process is performed as shown in FIGS. Then, in order to adaptively perform the quantization process for each pixel of the input image, the quantization error diffusion unit 7 transmits a signal based on the end of the process to the quantization threshold generation unit 3 after the end of the error diffusion process. (See FIG. 6).
[0080]
In the present embodiment, the first and second multi-level quantization processing by the quantization processing means 5 and the error diffusion processing by the quantization error diffusion means 7 are performed by a serpentine method as shown in FIG. Has become. That is, the first and second multi-level quantization processing and the error diffusion processing are not always performed in the same direction but are performed in the opposite direction by inverting each line. This makes it possible to easily disturb the uniform diffusion directions of the first and second quantization errors E1 and E2, thereby preventing the occurrence of "Worm" which is a drawback of the error diffusion processing. it can.
[0081]
In the above configuration, first, the pixel of the (i + 1) line near the pixel of interest (i, j) is used by using the feature quantity from the feature quantity extracting means 2, the first quantization threshold, and the first quantization level.(I + 1, j)Are subjected to quantization processing (first multi-level quantization processing) by the quantization processing means 5. Next, the first quantization error E1 generated in the process is output from the quantization error recognizing means 6 to the quantization error diffusing means 7, and the quantization error diffusing means 7 outputs a signal around the neighboring pixel (i + 1, j). (I + 1),i laIn pixel(I + 1, j-1), (i, j-1), (i, j), (i, j + 1)(Step 1). However, the pixels on the (i + 1) th line are quantized pixels(I + 1, j-1)Only the first quantization error E1 is diffused.
[0082]
Next, a line including the pixel of interest (i, j), that is, a pixel of the i-th line, using the feature amount, the second quantization threshold, and the second quantization level from the feature amount extraction unit 2.(I, j)Are subjected to quantization processing (second multi-level quantization processing) by the quantization processing means 5. Next, the second quantization error E2 generated in the process is output from the quantization error recognizing means 6 to the quantization error diffusion means 7, and the quantization error diffusion means 7 generates a second quantization error E2 around the pixel of interest (i, j). i, (i + 1) LAIn pixel(I, j + 1), (i + 1, j-1), (i + 1, j), (i + 1, j + 1)(Step 2). However, for pixels on the i-th line, pixels that have not been quantized(I, j + 1)Only the second quantization error E2 is diffused.
[0083]
The above steps 1 and 2 are regarded as one pair, and a series of processes are sequentially performed.To the processing direction of the arrow shown in FIG.Proceed as needed.
[0084]
According to the above configuration, after the quantization process and the error diffusion process on the (i + 1) line including the neighboring pixel (i + 1, j) are performed, the quantization process on the i line including the target pixel (i, j) is performed. Since the error diffusion processing is performed, it is possible to easily disturb the diffusion form of the first and second quantization errors E1 and E2. Therefore, even if the processed image is a gradation image having a gradual change in gradation, it is possible to avoid the occurrence of a stripe pattern called “Worm” unique to the multi-level error diffusion processing in a portion where the image density is high or a portion where the image density is low. And a good pseudo-tone reproduction image can be obtained.
[0085]
It is needless to say that the above-described second to fourth embodiments can be applied to the present embodiment, and the effect obtained by this is, of course, the same.
[0086]
【The invention's effect】
As described above, the image processing apparatus according to the first aspect of the present invention extracts the characteristic amount of the target pixel based on the pixel value of the target pixel to be processed and the neighboring pixels, and extracts the characteristic amount of the target pixel from the characteristic amount. A feature amount extracting unit that determines a region to which a pixel belongs; a quantization threshold generation unit that generates a quantization threshold value for performing a quantization process on the target pixel; and a quantization level for performing a quantization process on the target pixel. Quantization level generating means for generating, quantization processing means for quantizing the pixel value of the pixel of interest to a predetermined quantization level by comparing the pixel value of the pixel of interest with the quantization threshold, Means for defining a diffusion range of a quantization error generated in the quantization process by the means, determining a diffusion coefficient for each pixel within the diffusion range, and performing an error diffusion process. Conversion Value generating means and said quantization level generating means, in response to the regional-information of the pixel of interest, generating the quantization threshold value and the quantization levelFurther, when it is determined that the pixel of interest belongs to “photograph” or “indistinguishable area”, the position of the pixel of interest is set to (i , j), the quantization level generation means selects the combination of the quantization levels based on the remainder of (i + j) / 3.Configuration.
[0087]
Therefore, the same quantization level in the target pixel and its neighboring pixels is hardly continued, and the quantization process of the target pixel is adaptively performed according to the target pixel. It is possible to reduce the occurrence of texture caused by the same quantization level and the occurrence of false contours specific to multi-level error diffusion, even if the gradation image has a gradual change. This produces an effect that a tone reproduction image can be obtained.
[0088]
As described above, the image processing apparatus according to the second aspect of the present invention provides a target pixel to be processed.(I, j)And neighboring pixels(I + 1, j)A feature value extraction means for extracting a feature value of the pixel of interest based on the pixel value of the above, and determining a region to which the pixel of interest belongs based on the feature value; and a first quantization for quantizing the neighboring pixel. A threshold value generating means for generating a second quantization threshold value for performing a quantization process on the pixel of interest and a first quantization level for performing a quantization process on the neighboring pixels; A quantization level generating means for generating a second quantization level for performing a quantization process on the pixel of interest, and comparing the pixel value of the neighboring pixel with the first quantization threshold to obtain a pixel value of the neighboring pixel. Is quantized to a predetermined quantization level, and the pixel value of the pixel of interest is quantized to a predetermined quantization level by comparing the pixel value of the pixel of interest with the second quantization threshold. And the quantization hand Quantization error diffusion means for defining a diffusion range of a quantization error generated in the quantization process, determining a diffusion coefficient for each pixel within the diffusion range, and performing an error diffusion process. The processing meansPixel (i + 1, j) on line (i + 1)In addition to performing the first multi-level quantization processing, the quantization error diffusion unit performs the first multi-level quantization processing only on the pixels that have been subjected to the quantization processing for the pixels on the line including the neighboring pixels. Error diffusion processing for diffusing the first quantization error that occurs is performed, and thereafter, the quantization processing means includes:About the pixel (i, j) on the i-th lineIn addition to performing the second multi-level quantization processing, the quantization error diffusion unit performs the second multi-level quantization only on pixels that have not been subjected to the quantization processing with respect to the pixels on the line including the target pixel. This is a configuration for performing an error diffusion process for diffusing a second quantization error generated by the process.
[0089]
Therefore, after the quantization process and the error diffusion process on the line including the neighboring pixel are respectively performed, the quantization process and the error diffusion process on the line including the target pixel are performed, respectively. It is easily disturbed. Therefore, even if the processed image is a gradation image having a gradual change in gradation, it is possible to avoid the occurrence of a stripe pattern called “Worm”, which is unique to the multi-level error diffusion processing, in a portion where the image density is high or a portion where the image density is light. This has the effect that a good pseudo-tone reproduction image can be obtained.
[0090]
As described above, in the image processing apparatus according to the third aspect of the present invention, in the configuration of the second aspect, the quantization processing means and the quantization error diffusion means may perform the quantization processing and the error diffusion processing in a serpentine system. Respectively.
[0091]
Therefore, in addition to the effect of the configuration of claim 2, since the quantization process and the error diffusion process are not always performed in the same direction but are performed in the opposite direction by inverting each line, each quantization error Can be easily disturbed in the uniform diffusion direction. Therefore, there is an effect that "Worm" which is a defect of the error diffusion processing can be prevented from occurring.
[0092]
As described above, in the image processing apparatus according to the fourth aspect of the present invention, in the configuration of the second or third aspect, the quantization threshold value generation means and the quantization level generation means respond to the belonging area information of the target pixel. Thus, the second quantization threshold and the second quantization level are generated.
[0093]
Therefore, in addition to the effect of the configuration according to claim 2 or 3, the same quantization level does not almost continue in the target pixel and its neighboring pixels, and the quantization process of the target pixel is performed on the target pixel. Even if the processed image is a gradation image with a gradual change in gradation, the generation of textures due to the same quantization level continues and the multi-level error diffusion Can be reduced, and a good pseudo-tone reproduction image can be obtained.
[0094]
As described above, in the image processing apparatus according to the fifth aspect of the present invention, in the configuration according to any one of the first to fourth aspects, the quantization level generation unit may perform quantization of the quantized pixel near the pixel of interest. In this configuration, a quantization level of a target pixel to be quantized is generated based on the level.
[0095]
Therefore, in addition to the effect of any one of the first to fourth aspects, the probability that the quantization level of the target pixel and the quantization level of the already-quantized pixel are continuously the same is reliably reduced. Therefore, even if the pixel of interest belongs to, for example, a “photograph” in which a pseudo contour is particularly likely to occur, the probability that the pseudo contour occurs can be reliably reduced and a good quantized image can be reliably obtained. Play.
[0096]
As described above, in the image processing device according to the sixth aspect of the present invention, in the configuration of the fifth aspect, the quantization level generation unit generates a predicted quantization level based on a pixel value of the pixel of interest,When the predicted quantization level becomes the same as the intermediate quantization level,The quantization level is changed to a quantization level different from the predicted quantization level based on the magnitude relationship between the pixel value and the set quantization level.
[0097]
Therefore, in addition to the effect of the configuration of claim 5, the probability that the quantization level of the pixel of interest and its neighboring pixels will be continuously the same will be significantly reduced. As a result, the quantization error generated by the quantization process is also reduced. Therefore, even if the pixel of interest belongs to, for example, a "photograph" in which pseudo contours are particularly likely to occur, the probability of occurrence of pseudo contours can be greatly reduced, and a better quantized image can be reliably obtained. To play.
[0098]
As described above, in the image processing apparatus according to the seventh aspect of the present invention, in the configuration according to any one of the first to sixth aspects, the feature amount extracting means determines a region to which the pixel of interest belongs using a neural network. Configuration.
[0099]
Therefore, in addition to the effect of any one of the first to sixth aspects, the use of the neural network enables the region information of the target pixel to be accurately and accurately captured. Therefore, there is an effect that quantization processing adapted to each pixel can be performed with high accuracy and surely.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration example of an image processing apparatus according to the present invention.
FIG. 2 is an explanatory diagram showing an example of a diffusion range and a diffusion coefficient of a quantization error.
FIG. 3 is an explanatory diagram showing a neural network applied to the image processing apparatus.
FIG. 4 is an explanatory diagram showing another example of the quantization error diffusion range and the diffusion coefficient.
FIG. 5 is an explanatory diagram showing an example of a reference range of a quantization level.
FIG. 6 is a block diagram illustrating another configuration example of the image processing apparatus.
FIG. 7A is a diagram illustrating a state in which a first quantization error generated in a first multi-level quantization process is diffused, and FIG. 7B is a diagram illustrating a state in which the second multi-level quantization process is performed; FIG. 7 is an explanatory diagram showing a state in which a second quantization error generated by the diffusion is diffused.
FIG. 8 is an explanatory diagram showing an example of a serpentine method adopted in quantization processing and error diffusion processing in the image processing apparatus.
[Explanation of symbols]
2 Feature extraction means
3 Quantization threshold generation means
4 Quantization level generation means
5 Quantization processing means
7 Quantization error diffusion means

Claims (7)

In an image processing apparatus that converts input multi-tone image data into multi-valued quantized image data,
A feature amount extracting unit that extracts a feature amount of the noted pixel based on a pixel value of the noted pixel to be processed and its neighboring pixels, and determines a region to which the noted pixel belongs from the feature amount;
Quantization threshold generation means for generating a quantization threshold for performing quantization processing on the pixel of interest,
Quantization level generating means for generating a quantization level for performing quantization processing on the pixel of interest,
Quantization processing means for quantizing the pixel value of the target pixel to a predetermined quantization level by comparing the pixel value of the target pixel with the quantization threshold;
A quantization error diffusion unit that defines a diffusion range of a quantization error generated in the quantization process in the quantization unit, determines a diffusion coefficient for each pixel within the diffusion range, and performs an error diffusion process. ,
The quantization threshold generation means and the quantization level generation means generate the quantization threshold and the quantization level according to the region information of the target pixel,
Further, if it is determined that the pixel of interest belongs to a “photograph” or an “indistinguishable area”, and if the position of the pixel of interest is (i, j), the quantization level generating means will be (i + j) An image processing apparatus, wherein a combination of the quantization levels is selected based on a remainder of / 3. In an image processing apparatus that converts input multi-tone image data into multi-valued quantized image data,
Based on the pixel values of the pixel of interest (i, j) to be processed and its neighboring pixels (i + 1, j) , the characteristic amount of the pixel of interest is extracted, and the region to which the pixel of interest belongs is determined from the characteristic amount. Feature amount extracting means;
Quantization threshold generation means for generating a first quantization threshold for quantizing the neighboring pixels and generating a second quantization threshold for quantizing the pixel of interest;
Quantization level generation means for generating a first quantization level for quantizing the neighboring pixels and generating a second quantization level for quantizing the pixel of interest;
By comparing the pixel value of the neighboring pixel with the first quantization threshold, the pixel value of the neighboring pixel is quantized to a predetermined quantization level, and the pixel value of the target pixel and the second quantization threshold are calculated. By comparing the pixel value of the target pixel to a predetermined quantization level,
A quantization error diffusion unit that defines a diffusion range of a quantization error generated in the quantization process in the quantization unit, determines a diffusion coefficient for each pixel within the diffusion range, and performs an error diffusion process. ,
The quantization processing means performs the first multi-level quantization processing on the pixel (i + 1, j) of the (i + 1) line , and the quantization error diffusion means performs the first multi-level quantization processing on the pixel of the line including the neighboring pixel. Performing an error diffusion process for diffusing a first quantization error generated by the first multi-level quantization process only on pixels that have been subjected to the quantization process;
Thereafter, the quantization processing means performs the second multi-level quantization processing on the pixel (i, j) of the i-th line , and the quantization error diffusion means performs the second multi-level quantization processing on the pixel of the line including the target pixel. An image processing apparatus, comprising: performing an error diffusion process for diffusing a second quantization error generated by the second multi-level quantization process only on a pixel that has not been subjected to the quantization process. The image processing apparatus according to claim 2, wherein the quantization processing means and the quantization error diffusion means perform the quantization processing and the error diffusion processing, respectively, by a serpentine method. 3. The method according to claim 2, wherein the quantization threshold generation unit and the quantization level generation unit generate the second quantization threshold and the second quantization level according to the region information of the target pixel. Or the image processing apparatus according to 3. 5. The quantization level generating unit according to claim 1, wherein the quantization level generation unit generates a quantization level of the target pixel to be quantized based on a quantization level of a quantized pixel in the vicinity of the target pixel. An image processing device according to any one of the above. The quantization level generating means generates a prediction quantization level based on the pixel value of the target pixel, and sets the pixel value when the prediction quantization level becomes the same as the intermediate quantization level. The image processing apparatus according to claim 5, wherein the quantization level is changed to a quantization level different from the predicted quantization level based on a magnitude relationship with the quantization level. The image processing apparatus according to claim 1, wherein the feature amount extraction unit determines a region to which the pixel of interest belongs using a neural network.

Images