JP4010440B2 - Image processing apparatus with shadow detection - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
  The present invention relates to an image processing apparatus that processes an image read by an original reading apparatus, and more particularly to an image processing apparatus that includes shadow detection that detects and suppresses a shadow generated at a step portion of an original with high accuracy.
[0002]
[Prior art]
  2. Description of the Related Art Conventionally, in an image processing apparatus that reads a document with a scanner and performs contrast enhancement by filter processing, gamma conversion, or the like, “shadows” generated at a step portion of the document are also emphasized in the same manner as a character edge or an edge in a pattern. “Shadow” is a shadow of a cut and pasted document or a shadow of an edge when a thick paper is read with a white document holding plate as a background, and is actually unnecessary information, and thus becomes a problem.
[0003]
  Japanese Laid-Open Patent Publication No. 7-170405 is a technique that copes with the above-described shadow problem. Based on the density information, an edge shadow signal of the document is detected, and the detected edge shadow signal is removed from the image signal. However, it is impossible to distinguish from low-contrast characters by referring to density information. The low-contrast character is a low-density character typified by, for example, a pencil-written character. Naturally, although it is important as information, the method disclosed in JP-A-7-170405 also removes the low-contrast character part. This is also a problem.
[0004]
  Japanese Patent Application Laid-Open No. 10-285377 is another technique for dealing with the above-mentioned shadow defect. The shadow portion is determined based on the asymmetry of the edge gradient, and the shadow portion is deleted from the image signal. However, the detection based on the asymmetry of the edge gradient can be generally distinguished from the low-contrast character. However, since the character written in the pencil has a characteristic of asymmetry partially depending on the writing pressure, a false detection is inevitable. Further, shadow asymmetry is detected by referring to a read image obtained by light source irradiation from the left side and a read image obtained by light source irradiation on the right side, which is expensive as an apparatus.
[0005]
[Problems to be solved by the invention]
  An object of the present invention is to provide an apparatus that does not cause an unsightly shadow on an output image at low cost. For this reason, the shadow portion is identified as a low-contrast character and accurately detected, and control is performed so that the shadow portion is not enhanced or erased at least according to the shadow detection result.
[0006]
  In the present invention, in order to improve the shadow identification accuracy, a shadow candidate pixel is first detected and further corrected. In the example of Japanese Patent Laid-Open No. 10-285377, mention is made of correction using shadow continuity, but no specific method is disclosed. In the following embodiments, the correction means using the continuity of shadows and other properties will be specifically shown in a configuration that takes hardware into consideration.
[0007]
[Means for Solving the Problems]
  The invention of claim 1 is an image input means for acquiring an input image signal by irradiating a document;
A shadow detection means for detecting a shadow generated at a step portion in the thickness direction of the document from an input image signal;
Control means for controlling image processing on an input image signal based on a detection result of the shadow detection means, and the shadow detection means has background level detection means for detecting a background level, and the background level detection means If the density of the shadow candidate pixels detected by the shadow candidate pixel detecting means for detecting the shadow candidate pixels based on the background level step detected by the above and the shadow candidate pixel detecting means around the target pixel is greater than or equal to a predetermined threshold value And a correction unit that determines that the pixel is a pixel in a shadow area.
[0008]
  The invention of claim 2 is characterized in that, in the invention of claim 1, the control means controls the image processing so as to erase the input image signal of the shadow portion detected by the shadow detection means.
[0009]
  A third aspect of the invention is characterized in that, in the first aspect of the invention, the control means controls image processing so as not to at least enhance an input image signal of a shadow portion detected by the shadow detection means.
[0010]
  According to a fourth aspect of the present invention, in the first aspect of the invention, the shadow detecting unit includes a character feature detecting unit that detects a character feature from an input image signal, and the correcting unit is the pixel of interest or its surrounding pixels. If a character feature is detected by the detection means, it is determined that the target pixel is not a shadow region pixel.
[0011]
  According to a fifth aspect of the present invention, there is provided image input means for irradiating a document to obtain an input image signal, shadow detection means for detecting a shadow generated at a step portion in the thickness direction of the document from the input image signal, and the shadow detection means. Control means for controlling the image processing on the input image signal based on the detection result of the image, and the shadow detection means is detected by the shadow candidate pixel detection means for detecting the shadow candidate pixel, and the shadow candidate pixel detection means. The shadow candidate pixel detecting means has a background level detecting means for detecting a background level. The shadow candidate pixel detecting means has a background level detecting means for detecting a background level, and the shadow candidate pixel detecting means has a background level detected by the background level detecting means. A candidate pixel is detected, and the correction means refers to a shadow candidate pixel detection result by the shadow candidate pixel detection means around the target pixel, and the density of the shadow candidate pixels is equal to or greater than a predetermined threshold value If so, it is determined that the pixel of interest is a pixel in a shadow region, and the control means controls to erase or at least not enhance the input image signal of the shadow portion detected by the shadow detection means. Is.
[0012]
  According to a sixth aspect of the present invention, there is provided image input means for acquiring an input image signal by irradiating a document, shadow detection means for detecting a shadow generated at a step portion in the thickness direction of the document from the input image signal, and the shadow detection means. Control means for controlling the image processing on the input image signal based on the detection result of the image, and the shadow detection means is detected by the shadow candidate pixel detection means for detecting the shadow candidate pixel, and the shadow candidate pixel detection means. The shadow candidate pixel detecting means has a background level detecting means for detecting a background level. The shadow candidate pixel detecting means has a background level detecting means for detecting a background level, and the shadow candidate pixel detecting means has a background level detected by the background level detecting means. A candidate pixel is detected, and the correction unit refers to a shadow candidate pixel detection result by the shadow candidate pixel detection unit around the target pixel, and if the number of consecutive pixels of the shadow candidate pixel is equal to or greater than a predetermined threshold, Territory Is determined to be a pixel, the control means is characterized by controlling so or at least emphasized so as not to erase the input image signal of shadow detected by the shadow detection means.
[0013]
  According to a seventh aspect of the present invention, there is provided image input means for acquiring an input image signal by irradiating a document, shadow detection means for detecting a shadow generated at a step portion in the thickness direction of the document from the input image signal, and the shadow detection means. Control means for controlling image processing on the input image signal based on the detection result of the image, the shadow detection means is a shadow candidate pixel detection means for detecting a shadow candidate pixel, a character feature for detecting a pixel having a character feature Detection means, and correction means for correcting the shadow candidate pixels detected by the shadow candidate pixel detection means, wherein the shadow candidate pixel detection means includes background level detection means for detecting a background level, and the background level A shadow candidate pixel is detected based on the background level step detected by the detection means, and the character feature detection means detects a pixel having a character feature based on a two-dimensional density change specific to the character area, and The means refers to the character feature detection result by the character feature detection means around the target pixel, determines that the target pixel is not a shadow region pixel if the density of the pixels having the character feature is equal to or higher than a predetermined threshold, and the control The means controls the input image signal of the shadow portion detected by the shadow detection means to be erased or at least not emphasized.
[0014]
  According to an eighth aspect of the present invention, there is provided an image input means for irradiating a document to acquire an input image signal, a shadow detection means for detecting a shadow generated at a step portion in the thickness direction of the document from the input image signal, and the shadow detection means. Control means for controlling image processing on the input image signal based on the detection result of the image, the shadow detection means is a shadow candidate pixel detection means for detecting a shadow candidate pixel, a character feature for detecting a pixel having a character feature Detection means, and correction means for correcting the shadow candidate pixels detected by the shadow candidate pixel detection means, wherein the shadow candidate pixel detection means includes background level detection means for detecting a background level, and the background level A shadow candidate pixel is detected based on a background level step detected by the detection means, and the character feature detection means detects a pixel having a character feature based on a two-dimensional shape unique to the character area, and Refers to a character feature detection result by the character feature detection unit around the target pixel, determines that the target pixel is not a shadow region pixel if the density of pixels having the character feature is equal to or higher than a predetermined threshold, and the control unit Is characterized in that control is performed so as to erase or at least not enhance the input image signal of the shadow portion detected by the shadow detection means.
[0015]
DETAILED DESCRIPTION OF THE INVENTION
(Example)
  FIG. 1 is a block diagram forming the skeleton of the present invention. In the figure, 1 is a shadow detection unit, 2 is an image processing unit, a shadow detection unit 1 detects a shadow, and an image processing unit based on the detection result. 2 is controlled. As shown in FIG. 2, the shadow detection unit 1 includes a shadow candidate pixel detection unit 3 and a correction unit 4, but as shown in FIG. 3, a character feature detection unit 5 is provided to correct the character features. You may refer to it. Hereinafter, the shadow candidate pixel detection unit 3 in (1), the correction unit 4 in (2), the character feature detection unit 5 in (3), and the control of the image processing unit 2 based on the result of the shadow detection unit 1 in (4). Each will be described with some examples.
[0016]
  In each detection unit, for convenience of explanation and considering the hardware amount, binary determination or multi-value determination may be performed depending on circumstances, but it is possible to select binary output / multi-value output as necessary. It ’s fine. For example, in order to convert a binary output into a multi-value output, an appropriate smoothing filter may be applied to the binary output result, and the application is easy. Further, it may be performed as necessary in consideration of the relationship with the image processing system such as expansion processing for the determination result.
[0017]
(1) Embodiment of the shadow candidate pixel detection unit 3:
<Shadow candidate pixel detection using edge gradient>
  FIG. 4 is a diagram showing the relationship between the document step (FIG. 4A) and the pixel position and luminance in the vicinity of the document step (FIG. 4B), and the shadow is shown in FIG. 4B. In addition, the right and left ones have a steep slope edge b and the other is a gentle slope edge a. When the gradient is compared with the low contrast character image, the relationship is as shown in FIG. In FIG. 5, the steep gradient b has a larger gradient than the low contrast character image, and the gentle gradient a has a smaller gradient than the character image. Using this relationship, a shadow candidate pixel is detected.
[0018]
  FIG. 6 shows a processing flow. In FIG. 6, reference numeral 10 denotes an edge gradient detection filter, and 11 denotes a determination circuit. First, details of the edge gradient detection filter 10 will be described with reference to FIG. That is, in FIG. 7, the gradient is calculated from the pixel located across the pixel of interest at the center, and the gradient of the edge corresponding to the wide angle is obtained by calculating the gradient in four different directions and taking their maximum values. It becomes possible to detect. This can be described by an equation as follows.
X = max (| AB |, | CD |, | EF |, | GH |)
[0019]
  Next, the determination circuit 11 obtains the probability of being a shadow candidate pixel according to the gradient. When the scanner image is 600 dpi, the edge gradient detection filter 10 in FIG. 6 cannot detect the gradient b, which has a sharper shadow, because the resolution is insufficient. Therefore, as shown in FIG. 5, the probability of being a shadow candidate pixel can be obtained from the relationship between the gradient a having a gentle shadow and the gradient of the character image portion. Further, depending on the setting of the edge gradient detection filter 10, the gradient of b can be resolved. In this case, the determination may be made based on the relationship between the gradient of the shadow of b and the gradient of the character image portion.
[0020]
<Shadow candidate pixel detection using edge gradient asymmetry>
  As shown in FIG. 4, the shadow has the property that the edge gradient is asymmetric. Therefore, high detection accuracy can be expected by detecting the degree of gradient and further detecting asymmetry. In order to detect asymmetry, it is necessary to detect each of the shadow having the gradient of a and the shadow having the gradient of b in FIG. 5, and therefore, the gradient degree detection itself is required to have higher accuracy. Here, for the sake of simplicity of explanation, a hardware reduction in consideration of a practical aspect is considered, and the calculation is performed only in the main scanning direction. Of course, if this method is applied to several directions, high-accuracy detection can be achieved. In this respect, the method may be selected and implemented in consideration of cost performance.
[0021]
  FIG. 8 shows a processing flow. In FIG. 8, 20 is a primary differentiation circuit, 21 is a sign inverting circuit, 22 and 23 are maximum value detection circuits, 24 is a difference calculation circuit, 25 is a γ conversion circuit, and 26 is a determination circuit. The first differential value is obtained with reference to nine pixels g1 to g9 sandwiching the target pixel g5 in the main scanning direction. For example, the difference values s1 to s8 with adjacent pixels are set as the primary differential values. The relationship between the asymmetric gradient edge and the first derivative value is as shown in FIG. The absolute value max1 of the first derivative maximum value of the left edge with a steep slope is a large value, and conversely the absolute value max2 of the first derivative maximum value of the right edge with a gentle slope is a small value. Then, | max1-max2 | is calculated by the difference calculation circuit 24, and desired conversion is performed by the γ conversion circuit 25 (multi-value output in which the larger the difference value, the higher the probability of being a shadow candidate pixel). Alternatively, a binary determination result (a shadow candidate pixel [1] when the difference value is large, [0] otherwise) is output.
[0022]
  However, referring to nine pixels g1 to g9 as shown in FIG. 8, only a left or right edge gradient is detected even at a character edge having a certain thickness, so that it is regarded as an asymmetric edge. End up. Therefore, for example, based on a thin line detection result (a detection method will be described later), it is preferable that the comprehensive determination circuit 26 performs a determination that enables the determination of asymmetry only when the line is a thin line.
[0023]
<Shadow candidate pixel detection using directionality of edge gradient>
  In terms of the directionality of the edge gradient, the character area and the shadow area have the following characteristics. The character has a complicated shape, and the direction of the edge gradient in the local area is relatively different in terms of image data. On the other hand, the shadow area is noticeable when photographic paper is pasted. The direction of the edge gradient is linear. Here, a shadow candidate pixel is detected using this property.
[0024]
  FIG. 10 shows a processing flow. First, the edge gradient / direction detection filter 20 will be described in detail with reference to FIG. In other words, the gradient is calculated from the pixels located across the pixel of interest, and it is possible to detect the gradient of the edge more sensitively by calculating it in four different directions and taking their maximum values. . This can be described by an equation as follows.
[0025]
X = max (| AB |, | CD |, | EF |, | GH |)
First, X> Th20 (if not, output “0”),
| A-B | is the maximum of four, and if A> B, output "1"
If | A−B | is the maximum of four and A <B, output “2”
| C−D | is the maximum of four, and if C> D, output “3”
| C−D | is the maximum of four. If C <D, output “4”.
| E-F | is the maximum of four. If E> F, output "5"
| E−F | is the maximum of four, and if E <F, output “6”
| G-H | is the maximum of four, and if G> H, output "7"
| G-H | is the maximum of four, and if G <H, output "8"
[0026]
  Next, in the counting circuit 31, for example, in a mask of about 9 × 9, the values 1 to 8 are counted (the count values are Y1 to Y8, respectively). Then, in the comprehensive determination circuit 32, threshold values are determined for Y1 to Y8 with predetermined values. For example, when the count value (Yx) other than Y1 and Y2 exceeds the threshold value, the character region “1” is set, and not so. The case is determined and output as a shadow area “0”. Another character area determination case is as follows. Y3, Y4, and other values (Yx) exceed the threshold. Y5, Y6, and other values (Yx) exceed the threshold. Y7, Y8, and other values (Yx) exceed the threshold.
[0027]
<Shadow candidate pixel detection using edge boundary linearity>
  In terms of the linearity of the edge boundary, in general, a cut and pasted document is often rectangular, and the directionality of the edge boundary is often a straight line. Here, a shadow candidate pixel is detected using this property. However, as the calculation of the curvature, it is possible to convert the raster data of the edge portion of the image into vector data and calculate the curvature using, for example, Japanese Patent Application Laid-Open No. 63-298576 which is a conventional technique. When considering application to a copying machine or the like, the current cost performance is not necessarily smart in terms of processing amount and real-time performance, and a simple method with higher feasibility is selected here. FIG. 11 shows a processing flow.
[0028]
  In the processing flow of FIG. 11, the binarization circuit 40 is the first. The binarization circuit 40 has a low-contrast character and a shadow because the region of the image of interest is a portion such as a low-contrast character or shadow. Binarization with a fixed threshold value that is conscious of this is one means, but binarization with a flow-type threshold value, that is, an average value is calculated in a predetermined mask area including the target pixel, and basically Features can be extracted more sensitively by binarization based on the values. The pattern matching circuit 41 detects linearity with respect to the binarization result.
[0029]
  FIG. 12 shows an example of pattern matching in the horizontal direction. Hereinafter, the detection of the linearity in the horizontal direction with respect to the edge will be described as an example, but application to the vertical direction and the tanning direction is easy. When the pixel of interest matches FIG. 12, the pixel of interest is “1”, and when it does not match, it is “0”.
[0030]
<Detection of shadow candidate pixels using background detection>
  The background area of characters is typically a white background, but in any case, it is often uniform. As shown in FIG. 13, the shadow area of cut and pasted is the ground level of the pasted paper and the pasted paper. Often the ground level is different. Here, a shadow candidate pixel is detected using the feature. In addition, for the sake of simplicity of explanation, a reduction in hardware in consideration of practical aspects is taken into consideration, and here, the calculation is performed only in the main scanning direction. Of course, if this method is applied to several directions, the detection becomes highly accurate. In this respect, the method may be selected and implemented in consideration of cost performance.
[0031]
  FIG. 14 shows a processing flow. In the average value calculation circuit 50, the pixel of interest X as shown in FIG. 15 is sandwiched between two pixels (A50, B50) separated by a predetermined distance in the main scanning direction (for example, 7 × 7 pixels). On the other hand, an average value (n50, n51) is obtained. Then, | n50−n51 | is calculated by the difference calculation circuit 51, and further, the value is subjected to desired conversion by the γ conversion circuit 52 (the larger the difference value, the higher the probability of being a shadow candidate pixel). Multi-value output) or binary determination result (when the difference value is large, the shadow candidate pixel [1], otherwise [0]) is output. By the way, as the average value (n50, n51) is calculated with a larger mask, calculation with higher accuracy becomes possible. On the other hand, there is a greater possibility that information other than the background, for example, a character part or the like is included. Therefore, when obtaining an average value, regarding a pixel value having a density value higher than a predetermined value, means for rejecting the value is also very effective. In addition, this time, the image data read by a scanner is used to determine whether the background of the edge area is made of uniform paper or multiple papers. For example, the glossiness of the paper surface is measured. It is also possible to automatically recognize paper using a sensor (for example, manufactured by Nichicon). In the future, it is also possible to recognize a single paper / multiple paper in the vicinity of a predetermined edge by using a plurality of such sensors.
[0032]
<Shadow candidate pixel detection using line width detection>
  As can be seen from FIG. 13, both the edge of the character and the edge of the shadow are made up of paired edges whose gradient directions are opposite to each other. The width of the paired edge has a property that the shadow part is relatively narrow compared to the width of the paired edge of a pencil-written character that is generally used as a low-contrast character on an actual image. Here, a shadow candidate pixel is detected by thin line detection using the above-described properties. In addition, for the sake of simplicity of explanation, and considering a reduction in hardware in consideration of practical aspects, the calculation is performed only in the main scanning direction. Of course, if it is carried out in several directions, it becomes highly accurate, and in this respect, the method may be selected and implemented in consideration of cost performance.
[0033]
  FIG. 16 shows a processing flow. The line thickness calculation circuit 60 obtains the character width and shadow width of the target pixel. The calculation method here uses, for example, the character width calculation of the character width determination unit described in Japanese Patent Application Laid-Open No. 8-130636, which is a conventional device. Then, a desired conversion is performed by the γ conversion circuit 61 (multi-value output such that the smaller the line thickness is, the higher the probability of being a shadow candidate pixel), or the binary determination result (the line thickness is small) In this case, the shadow candidate pixel [1] is output. Otherwise, [0]) is output. The above is a description of a specific example of the shadow candidate pixel detection 3.
[0034]
(2) Example of correction unit 4:
<When correcting by referring to the peripheral shadow candidate pixel detection result>
  << Density >> The shadow candidate pixels are detected using a feature that prevents erroneous detection of low-contrast characters. For example, in shadow candidate pixel detection using the asymmetry of the edge gradient, pencil writing is used. If the writing pressure is biased by characters, false detections occur in some places. Therefore, it is very effective to perform density correction with reference to the shadow candidate pixel detection result of the surrounding pixels. Of course, the present invention is not limited to this, and is also effective for detecting other shadow candidate pixels.
[0035]
  The processing flow is shown in FIG. First, the average value calculation circuit 70 refers to, for example, 9 × 9 pixel shadow candidate pixel detection results centered on the target pixel, and obtains the average value. In the case of binary determination of [0/1] as an output of shadow candidate pixel detection, the count of shadow candidate pixels corresponds to that. Then, the γ conversion circuit 71 performs conversion so as to obtain a desired determination result (multi-level output such that the higher the density, the higher the probability of being a shadow). Alternatively, binary determination is performed using a threshold value (shadow area [1] when the density is high, and [0] otherwise).
[0036]
  << Continuity >> The edge of a cut and pasted document, or the edge when a thick paper is read with a white document holding plate as a background is a closed region, and the shadow formed there is also a closed line. Therefore, the shadow has a property of continuity, and here, the shadow candidate pixel is corrected using the property. For example, when shadow candidate pixel detection using shadow linearity is performed, lines having linearity are also present in characters, and are detected in the shadow candidate pixel detection result stage. However, characters and shadows differ greatly in terms of continuity, and in the case of characters, even if they are straight, they have the property of being short, so this correction is very effective. Of course, the present invention is not limited to this, and is also effective for detecting other shadow candidate pixels.
[0037]
  FIG. 18 shows a processing flow. The continuous pixel number counting circuit 80 counts up the number of continuous pixels from various directions, diagonally and vertically. This will be specifically described with reference to FIG. When the shadow candidate pixel detection result of the pixel of interest f is a non-shadow candidate pixel of 0 (here, for the sake of simplicity, it is assumed that binary determination is performed), the number of consecutive pixels at the pixel of interest n (f) = 0 And store in memory. When the shadow candidate pixel detection result of the target pixel f is 1, the continuous pixel number n of the pixels at positions a, b, c, d, and e corresponding to the previous pixel in the main and sub scanning directions with respect to f. (A), n (b), n (c), n (d), n (e) are read from the image memory 81 and n (f) = max (n (a), n (b), n (c) ), N (d), n (e)) + 1. The binarization circuit 82 outputs “1” as a shadow region if the number of continuous pixels at the target pixel position is greater than or equal to a predetermined value, and outputs “0” otherwise. Further, the shadow area is expanded by the expansion circuit 83. For example, if there is at least one “1” in a 5 × 5 mask including the target pixel, a process for setting the target pixel to “1” is performed.
[0038]
  << Uniformity >> When the shadow candidate pixel detection is multi-valued determination, the shadow has a property of uniformity of the multi-valued determination result. For example, in shadow candidate pixel detection using line width, a thin line is detected, but if this is output in multiple values, the line width of the shadow portion is uniform and that of the character portion is non-uniform compared to the shadow portion. Can be used to correct. Not only this, but also in shadow candidate pixel detection using other features, the shadow edge has the feature that the feature is almost uniform for the original with the same thickness and the shadow in the same direction. Correction using is effective. Shadow candidate pixel detection using the directionality of the edge gradient originally uses the uniformity of the gradient direction, but it may be corrected by the uniformity by referring to a wider area. In addition, for the sake of simplicity of explanation, a reduction in hardware in consideration of practical aspects is taken into consideration, and here, the calculation is performed only in the main scanning direction. Of course, if it is carried out in several directions, it becomes highly accurate, and in this respect, the method may be selected and implemented in consideration of cost performance.
[0039]
  FIG. 20 shows a processing flow. First, the shadow candidate pixel extraction circuit 90 extracts the shadow candidate pixel detection results KA and KB of two pixels A and B separated by a predetermined distance in the main scanning direction with the target pixel interposed therebetween. This is the same as FIG. Alternatively, for example, an average value of the shadow candidate pixel detection results may be calculated with reference to the peripheral pixel region of the pixel A and the pixel B. Then, | KA−KB | is calculated by the difference calculation circuit 91, and further, the value is subjected to desired conversion by the γ conversion circuit 92 (multi-value such that the smaller the difference value, the higher the probability of being a shadow. Output) or binary determination result (shadow area [1] when the difference value is small, [0] otherwise). Of course, the result of the γ conversion circuit 92 may be expanded as necessary in consideration of the relationship with the image processing system as before.
[0040]
<When correcting by referring to the surrounding character detection result>
  Heretofore, several methods for correcting with reference to the shadow candidate pixel detection results have been shown. In addition to this, as shown in FIG. 3, there is a method in which character feature detection 5 is performed and the result is also referred to for correction. In contrast, the low-contrast character portion erroneously detected as the shadow candidate pixel is removed by detecting the feature of the character portion. When it is detected as a shadow candidate pixel and the pixel of interest or the surrounding pixel is detected as a character in the character feature detection, the character feature detection result is given priority, and the result of the correction unit 4 is set as a non-shadow. This is because the character information is naturally more useful information and should not affect the image quality.
[0041]
  《Correction with density of surrounding character feature detection result》 The probability that there is a character around the character is reasonably high, that is, the character edge density is generally higher than the shadow edge density. . It is also effective to perform correction using the property that “there is a character around the character”.
[0042]
  FIG. 21 shows a processing flow. Here, for simplicity of explanation, both of the shadow candidate pixel detection result and the character feature detection result are binary determinations. In the density calculation circuit 100, for example, the density is obtained by referring to the character feature detection result for 9 × 9 pixels in the periphery. What is necessary is just to carry out similarly to the correction circuit by the density of FIG. Here, for simplicity of explanation, the output from the density calculation circuit 100 is a binary output (1 is high density, 0 is low density). The shadow candidate pixel detection result is also a binary output (1 is a shadow candidate, 0 is a non-shadow candidate). Then, a shadow pixel is determined by the determination circuit 101 from these two results. Similarly to the above, when the character feature detection is high density: 1, it is determined as non-shadow: 0. When the character feature detection is low density: 0, the shadow candidate pixel detection result is used as the determination result of the shadow determination. Of course, when combined with density correction of shadow candidate pixels and correction using continuity, correction and detection with higher accuracy can be performed. The above is the description of a specific embodiment of the correction unit 4.
[0043]
(3) Example of character feature detection unit 5:
<Character detection using two-dimensional density detection unique to character areas>
  The character region often has a characteristic two-dimensional density changing portion. Here, for the multi-value data, those local patterns are detected, and pixels having character characteristics are detected. FIG. 22 shows a processing flow. Some specific examples of the two-dimensional density change detection circuit 110 will be described below.
a) Example of a two-dimensional density change detection circuit using a Laplacian filter
The character area includes many crossed portions such as the center of “x”. On the other hand, the shadow area is hardly included. Utilizing this feature, for example, a Laplacian filter as shown in FIG. 23 is applied to the image data, and if the filter output value is equal to or greater than a predetermined value, the target pixel is set as a pixel region having a large two-dimensional density change.
[0044]
b) Example of a two-dimensional density change detection circuit using a peak pixel detection filter
Using the features described above, for example, the peak pixel detection described in the paper “Image Area Separation Method for Mixed Character / Picture (Half Dots and Photos) Images” may be used. However, although it is described in the paper as a process for detecting halftone dots, it is also a response to the cross area of the character part. Of course, the density of the peak pixels in the character portion and the halftone dot portion is larger than the comparison, and the fact that the character portion is detected is not a story that changes the conventional gist.
[0045]
c) Example of a two-dimensional density change detection circuit using a ridge pixel detection filter
Both the character area and the shadow portion of the cut and pasted document have a ridge shape. Furthermore, the symmetry and asymmetry of the mountain is further improved by taking advantage of the fact that the background of the character part is, for example, a uniform white background, while the background part of the shadow part may change when cut and pasted. The two are separated by use (see FIG. 13). Conventionally, ridge pixel detection is described in detail in Japanese Patent Laid-Open No. 3-82269, and here, a condition for adding symmetry of the density gradient is added to the conventional processing. An example of a filter for detecting the symmetry of the density gradient is shown in FIG.
[0046]
d) Example of a two-dimensional density change detection circuit using a two-dimensional black-and-white change detection process
The frequency of density change in the character area is large in the predetermined area, while that in the shadow area is small. Here, after binarization by a predetermined method (the flow type threshold as described above is better than binarization), the binarized data is subjected to each of the main and sub masks in a mask having a two-dimensional spread. If the number of change points from white to black or black to white is counted in the scanning direction and the counted value is greater than or equal to a predetermined value, the pixel of interest in the mask or all the pixels in the mask have a large two-dimensional density change. Let it be a pixel area.
[0047]
<Character detection using two-dimensional shape detection unique to character regions>
  The character region often has a characteristic two-dimensional shape. Here, a local pattern unique to the character part is detected for the data after binarization processing or for the ridge pixel detection result as described in the conventional Japanese Patent Laid-Open No. 3-82269, A pixel having a character feature is detected. FIG. 25 shows a processing flow. First, although the binarization circuit 120 is used, binarization may be performed by a predetermined method (the binarization is better with a flow-type threshold as described above), or ridge pixel detection may be performed. . Then, the pattern matching circuit 121 detects a characteristic shape existing in the character portion described in detail below.
[0048]
a) Detection of character crossing part by pattern matching: Needless to say, the character part has many cross parts, while the shadow part has little. Using this feature, a pattern as shown in FIG. 26 is prepared and detected by pattern matching. Of course, it is effective to prepare a pattern such as “×”.
[0049]
b) Detecting the “T” part of the character by pattern matching: Needless to say, the character part has many intersections such as “T”, while the shadow part has little. Using this feature, a pattern as shown in FIG. 27 is prepared and detected by pattern matching. Of course, it is also effective to prepare a pattern that rotates this.
[0050]
c) Detection of the end of the character by pattern matching: Needless to say, the character has many ends, while the shadow has little. Using this feature, a pattern as shown in FIG. 28 is prepared and detected by pattern matching. Of course, it is also effective to prepare a pattern that rotates this. The above is the description of a specific embodiment of the character feature detection unit 5.
[0051]
(4) Example of control of image processing 2 based on the result of the shadow detection unit 1: For the sake of simplicity, the shadow detection result is a binary output. It is also possible to expand.
<Eliminate the shadow>
  In response to the shadow detection result, the shadow edge is erased. FIG. 29 is a processing flow of shadow edge elimination. In the selector 130, when the shadow detection result is 1, the signal a in which the shadow edge is eliminated by the minimum value detection unit 131 in the M × N pixel is selected, and when the shadow detection result is 0, the through signal b is selected by the through unit 132. ,Output. The signal a from which the shadow edge is erased is the minimum density value of the M × N pixels around the target pixel, and the shadow is erased by replacing the target pixel value with this value.
[0052]
<Do not emphasize shadows at least>
  A control method that does not completely eliminate the shadow and at least does not emphasize the shadow is also very effective. Two examples are given below: an example in which shadow edges are suppressed by filter processing control, and an example in which a similar object is achieved by gamma conversion control. Compared with an image in which a conventional shadow portion has been enhanced by filtering or gamma conversion, there is a sufficient effect of eliminating an obstructive level of shadow edges on the output image.
[0053]
  FIG. 30 is a processing flow of shadow edge suppression by filter processing control. The edge emphasis unit 133 performs edge emphasis on the input image signal read by the scanner, and sends a signal a whose contrast is enhanced at the edge part to the selector 130. At the same time, a through signal b that does not pass through the edge enhancement unit 133 is also sent to the selector 130. The selector 130 receives the result of the shadow detection and receives the shadow detection result.Is 0Output signal a after edge enhancement if any, shadow detection resultIs 1If there is, the through signal b is selected and output. If not only the edge enhancement filter but also the smoothing filter is controlled, a shadow edge suppression effect can be obtained.
[0054]
  FIG. 31 is a processing flow for shadow edge suppression by gamma conversion control. The gamma conversion unit 1 indicated by 134 performs gamma conversion on the input image signal read by the scanner, for example, using the LUT for gamma conversion 1 shown in FIG. 32, and sends the converted signal a to the selector 130. Similarly, the gamma conversion unit 2 indicated by 135 performs gamma conversion using the LUT for gamma conversion 2 in FIG. 32, and sends the converted signal b to the selector 130. The selector 130 receives the result of the shadow detection and receives the shadow detection result.Is 0Output signal a after gamma conversion 1 if there is, shadow detection resultIs 1If there is, the signal b after the gamma conversion 2 is selected and output. If the LUT for gamma conversion 2 shown in FIG. 32 has a downwardly convex shape, it is possible to suppress the signal of the shadow portion.
[0055]
【The invention's effect】
  According to the present invention, when a shadow region is detected from an input image, a shadow candidate pixel is first detected and then corrected, so that a two-stage configuration is employed. The shadow area can be detected with high accuracy while suppressing the detection. Further, since the image processing is controlled using the shadow detection result, an unsightly shadow can be suppressed.
[0056]
  By erasing the shadow area from the input image, it is possible to eliminate lines generated in the shadow portion.
[0057]
  By not emphasizing the shadow edge at least, the line generated in the shadow portion can be set to a level that is hardly noticed on the output image.
[0058]
  Since shadow candidate pixel detection based on local information is performed, an apparatus with good cost performance can be put into practical use.
[0059]
  Since the shadow candidate pixel detection result of the target pixel is corrected with reference to the surrounding shadow candidate pixel detection results, the detection accuracy that has been insufficient in the past can be increased to a practical level. In addition, since the correction is made in consideration of the hardware amount, an apparatus with good cost performance can be put into practical use.
[0060]
  Since pixels having character characteristics are detected and the shadow candidate pixel detection result of the target pixel is corrected with reference to this, the detection accuracy that has been insufficient in the past can be increased to a practical level. In addition, since the correction is made in consideration of the hardware amount, an apparatus with good cost performance can be put into practical use.
[0061]
  Since character feature detection based on local information is performed, an apparatus with good cost performance can be put into practical use.
[Brief description of the drawings]
FIG. 1 is a block diagram of a main part forming a skeleton of the present invention.
FIG. 2 is a diagram for explaining an example of a shadow detection unit 1 shown in FIG. 1;
FIG. 3 is a diagram for explaining another example of the shadow detection unit 1 shown in FIG. 1;
FIG. 4 is a diagram illustrating a state of a document step portion and a relationship between a pixel position near the step portion and luminance.
FIG. 5 is a diagram illustrating the relationship between the gradient of an edge and the probability of being a shadow.
FIG. 6 is a block diagram illustrating an example of shadow detection using an edge gradient degree.
FIG. 7 is a diagram illustrating an example of a filter for detecting an edge gradient.
FIG. 8 is a diagram illustrating an example of shadow detection using edge gradient asymmetry.
FIG. 9 is a diagram showing a relationship between an asymmetric gradient edge and a first derivative value.
FIG. 10 is a principal block diagram for explaining an example of shadow detection using the directionality of an edge gradient.
FIG. 11 is a principal block diagram for explaining an example of shadow detection using the linearity of an edge boundary;
FIG. 12 is a diagram showing an example of pattern etching.
FIG. 13 is a diagram illustrating a relationship between a background and a density of a shadow portion.
FIG. 14 is a block diagram of relevant parts for explaining an example of shadow detection using detection of a back ground.
15 is a diagram for explaining an operation example of an average value equalizing circuit shown in FIG. 14; FIG.
FIG. 16 is a principal block diagram for explaining an example of shadow detection using line width detection;
FIG. 17 is a diagram for explaining an example of pixel correction using peripheral pixel density;
FIG. 18 is a diagram for explaining an example of pixel correction using a continuous line of shadows.
FIG. 19 is a diagram for explaining an operation side of the continuous pixel number counting circuit shown in FIG. 18;
FIG. 20 is a diagram for explaining an example of pixel correction using a uniform width of shadow pixels.
FIG. 21 is a diagram for explaining an example of pixel correction using the characteristics of surrounding characters.
FIG. 22 is a diagram for explaining an example of a character feature detection unit 5 shown in FIG. 3;
FIG. 23 is a diagram for explaining an example of detecting a two-dimensional density using a Laplacian filter.
FIG. 24 is a diagram illustrating an example of a filter that detects the symmetry of a density gradient.
FIG. 25 is a diagram illustrating an example of character detection using two-dimensional shape detection unique to a character region.
FIG. 26 is a diagram illustrating an example of a character cross portion detection pattern.
FIG. 27 is a diagram illustrating an example of a pattern for detecting a T portion of a character.
FIG. 28 is a diagram illustrating an example of a pattern for detecting an end portion of a character.
FIG. 29 is a diagram for explaining an example in which a shadow edge is deleted using a minimum density value;
FIG. 30 is a diagram for explaining an example of shadow edge control by filter processing;
FIG. 31 is a diagram for explaining an example of shadow edge control by gamma conversion control;
FIG. 32 is a diagram for explaining an operation of shadow edge control using gamma conversion.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Shadow detection part, 2 ... Image processing part, 3 ... Shadow candidate pixel detection part, 4 ... Correction | amendment part,
5 ... Character feature detection unit, 10 ... Edge gradient detection filter, 11 ... Determination circuit,
20 ... primary differentiation circuit, 21 ... sign inversion circuit, 22, 23 ... maximum value detection circuit,
24 ... difference calculation circuit, 25 ... gamma conversion circuit, 26 ... determination circuit,
30: Edge gradient / direction detection filter, 31: Count circuit, 32: Comprehensive determination circuit,
40 ... binarization circuit, 41 ... pattern matching circuit, 50 ... average value calculation circuit,
51 ... Difference calculation circuit, 52 ... γ conversion circuit, 60 ... Line thickness calculation circuit, 61 ... γ conversion circuit, 70 ... Average value calculation circuit, 71 ... γ conversion circuit, 80 ... Continuous pixel number counting circuit,
81: Image memory, 82: Binarization circuit, 83: Expansion circuit, 90: Shadow candidate pixel extraction circuit,
91 ... Difference calculation circuit, 92 ... γ conversion circuit, 100 ... Density calculation circuit, 101 ... Determination circuit,
110 ... Two-dimensional density change detection circuit, 111 ... Density correction circuit, 112 ... Expansion circuit,
120: binarization circuit, 121: pattern matching circuit, 122: expansion circuit,
130... Selector, 131... Minimum value detection unit in M × N pixels, 132.
133: Edge emphasis unit, 134, 135: Gamma conversion unit.

Claims (8)

Image input means for irradiating a document to obtain an input image signal;
Shadow detection means for detecting a shadow generated at a step portion in the thickness direction of the document from an input image signal;
Control means for controlling image processing on the input image signal based on the detection result of the shadow detection means,
The shadow detection means includes
A shadow candidate pixel detecting means for detecting a shadow candidate pixel based on a step of the background level detected by the background level detecting means;
Correction means for determining that a pixel of interest is a pixel in a shadow region if the density of the shadow candidate pixels detected by the shadow candidate pixel detection unit around the pixel of interest is equal to or higher than a predetermined threshold value. An image processing apparatus.   The image processing apparatus according to claim 1, wherein the control unit controls image processing so as to erase the input image signal of the shadow portion detected by the shadow detection unit.   The image processing apparatus according to claim 1, wherein the control unit controls image processing so as not to enhance at least an input image signal of a shadow portion detected by the shadow detection unit. The shadow detection means includes character feature detection means for detecting a character feature from an input image signal, and the correction means is focused if a character feature is detected by the character feature detection means at the target pixel or its surrounding pixels. The image processing apparatus according to claim 1, wherein the pixel is determined not to be a pixel in a shadow area. Image input means for irradiating a document and obtaining an input image signal;
  A shadow detection means for detecting a shadow generated at a step portion in the thickness direction of the document from an input image signal;
  Control means for controlling image processing on the input image signal based on the detection result of the shadow detection means,
  The shadow detection means includes a shadow candidate pixel detection means for detecting a shadow candidate pixel, and a correction means for correcting the shadow candidate pixel detected by the shadow candidate pixel detection means,
  The shadow candidate pixel detection means has a background level detection means for detecting a background level, detects a shadow candidate pixel based on a step of the background level detected by the background level detection means,
  The correction means refers to the shadow candidate pixel detection result by the shadow candidate pixel detection means around the target pixel, and determines that the target pixel is a shadow area pixel if the density of the shadow candidate pixels is equal to or higher than a predetermined threshold. ,
  The image processing apparatus according to claim 1, wherein the control unit performs control so as to erase or at least not enhance an input image signal of a shadow portion detected by the shadow detection unit. Image input means for irradiating a document and obtaining an input image signal;
  A shadow detection means for detecting a shadow generated at a step portion in the thickness direction of the document from an input image signal;
  Control means for controlling image processing on the input image signal based on the detection result of the shadow detection means,
  The shadow detection means includes a shadow candidate pixel detection means for detecting a shadow candidate pixel, and a correction means for correcting the shadow candidate pixel detected by the shadow candidate pixel detection means,
  The shadow candidate pixel detection means has a background level detection means for detecting a background level, detects a shadow candidate pixel based on a step of the background level detected by the background level detection means,
  The correction means refers to the shadow candidate pixel detection result by the shadow candidate pixel detection means around the target pixel, and if the number of continuous pixels of the shadow candidate pixel is equal to or larger than a predetermined threshold, the target pixel is a shadow area pixel. Judgment,
  The image processing apparatus according to claim 1, wherein the control unit performs control so as to erase or at least not enhance an input image signal of a shadow portion detected by the shadow detection unit. Image input means for irradiating a document and obtaining an input image signal;
  A shadow detection means for detecting a shadow generated at a step portion in the thickness direction of the document from an input image signal;
  Control means for controlling image processing on the input image signal based on the detection result of the shadow detection means,
  The shadow detection means includes a shadow candidate pixel detection means for detecting a shadow candidate pixel, a character feature detection means for detecting a pixel having a character feature, and a correction for correcting the shadow candidate pixel detected by the shadow candidate pixel detection means. With means,
  The shadow candidate pixel detection means has a background level detection means for detecting a background level, detects a shadow candidate pixel based on a step of the background level detected by the background level detection means,
  The character feature detection means detects a pixel having a character feature based on a two-dimensional density change specific to the character region,
  The correction unit refers to the character feature detection result by the character feature detection unit around the target pixel, determines that the target pixel is not a shadow region pixel if the density of the pixels having the character feature is equal to or higher than a predetermined threshold,
  The image processing apparatus according to claim 1, wherein the control unit performs control so as to erase or at least not enhance an input image signal of a shadow portion detected by the shadow detection unit. Image input means for irradiating a document and obtaining an input image signal;
  A shadow detection means for detecting a shadow generated at a step portion in the thickness direction of the document from an input image signal;
  Control means for controlling image processing on the input image signal based on the detection result of the shadow detection means,
  The shadow detection means includes a shadow candidate pixel detection means for detecting a shadow candidate pixel, a character feature detection means for detecting a pixel having a character feature, and a correction for correcting the shadow candidate pixel detected by the shadow candidate pixel detection means. With means,
  The shadow candidate pixel detection means has a background level detection means for detecting a background level, detects a shadow candidate pixel based on a step of the background level detected by the background level detection means,
  The character feature detection means detects a pixel having a character feature based on a two-dimensional shape unique to the character region,
  The correction unit refers to the character feature detection result by the character feature detection unit around the target pixel, determines that the target pixel is not a shadow region pixel if the density of the pixels having the character feature is equal to or higher than a predetermined threshold,
  The image processing apparatus according to claim 1, wherein the control unit performs control so as to erase or at least not enhance an input image signal of a shadow portion detected by the shadow detection unit.

Images