JPH027553B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention provides a motion correction noise method that reduces noise in an image signal by using correlation between frames while correcting the image position corresponding to the direction and size of the movement of the image. The present invention relates to a noise reduction device that specifically reduces noise in such a way that malfunctions due to the noise itself do not occur.

In this type of motion compensation noise reduction device, as described in Japanese Unexamined Patent Publication No. 55-37018 filed by the applicant, a frame memory is used to positively feed an image signal delayed by a frame to an input image signal. When reducing noise in the output image signal by adding and emphasizing image signal components by inter-frame correlation and averaging noise with no inter-frame correlation, the inter-frame correlation of the images is lost due to image movement. Motion correction reduces image signal components along with noise based on the results of detecting image movement, and performs position correction on image signal components for areas where there is movement in the image to restore inter-frame correlation. By giving positive feedback to the predicted image signal, it is possible to reduce only the noise in the image signal of the moving part.The presence or absence of image movement is detected for each block in an appropriate range, and the detection result is used to reduce the noise. The motion-corrected predicted image signal and the non-motion-corrected predicted image signal are switched in block units to provide positive feedback. However, since the presence or absence of image movement is detected based on the size of the difference between image signal frames, when a large difference due to noise occurs between frames, motion correction is also applied to the image signal of the part where there is no movement. Since the manipulated motion-corrected predicted image signal is added by positive feedback, there is a drawback that the image quality deteriorates.

An object of the present invention is to eliminate the above-mentioned drawbacks of the conventional art, and to apply positive feedback switching between a motion-corrected predicted image signal and a non-motion-corrected predicted image signal, which is performed for each pixel according to the detection result of the presence or absence of image movement. An object of the present invention is to provide a motion compensation noise reduction device that prevents malfunctions due to the influence of noise during detection.

That is, the motion compensation noise reduction device of the present invention
A movement of reducing noise in the input image signal by inter-frame correlation between the input image signal and a predicted image signal in which the position of the image is corrected according to the direction and size of the movement of the input image signal one frame before. In the correction noise reduction device, for each successive pixel in the input image signal, image movement is detected for a plurality of pixels including pixels adjacent to each pixel, and the image movement is detected according to the direction and magnitude of the image movement. Based on the results of comparing the motion-compensated predicted image signal subjected to position correction and the non-motion-compensated predicted image signal without position correction with the input image signal to determine their respective approximations,
The present invention is characterized in that the motion-compensated predicted image signal and the non-motion-compensated predicted image signal are switched for each successive pixel to produce the predicted image signal.

The invention will be explained in detail below by way of example embodiments with reference to the drawings.

Generally, in a motion compensation noise reduction device, an image signal predicted by applying positional correction for image movement to an image signal of the previous frame based on the difference between frames is very similar to the image signal of the current frame. However, in the past, position correction for movement was performed in units of blocks, such as 8 lines x 16 pixels, so if the same block contains moving pixels and stationary pixels, When looking at the motion-corrected predicted image signal pixel by pixel, it is not necessarily very similar to the image signal of the current frame, and as a result of applying motion correction, still pixels move along with moving pixels. A new difference will arise between the frame image signal and the frame image signal. Therefore, between the motion-compensated predicted image signal and the non-motion-compensated predicted image signal, the one that most closely approximates the current frame image signal may be used for noise reduction. In short, the algorithm uses a motion-compensated predicted image signal or a non-motion-compensated predicted image signal, whichever approximates the current frame image signal. On the other hand, the image signal that should undergo noise reduction is
A noisy image signal with a reduced signal-to-noise ratio, so significant noise may be interpreted as image movement, and switching may not occur completely. .

Therefore, switching between a motion-compensated predicted image signal and a non-motion-compensated predicted image signal based on the result of simply detecting the presence or absence of motion for each pixel, as in the past, does not provide a sufficient noise reduction effect. Not only is this not possible, but there is even the possibility that switching will cause a new increase in noise.

On the other hand, conventional motion detection was performed for each block of the image, and the noise within each block was averaged to produce a reduction effect, so the motion detection itself was performed accurately. It can be considered as a thing. However, the switching signal resulting from the detection was not necessarily accurate for the reasons mentioned above. Furthermore, if the above-mentioned switching is performed in units of blocks of an image, distortion occurs between blocks, which is not appropriate. Therefore, in the device of the present invention, first, switching between a motion-compensated predicted image signal and a non-motion-compensated predicted image signal is performed pixel by pixel, and the presence or absence of movement in the image, which is the basis for the switching, is detected for each image. In addition, the motion detection performed for each image is not simply performed on successive pixels, but is comprehensively performed on each successive pixel, including surrounding pixels, and motion detection is performed only on each individual pixel. This is to prevent the above-mentioned erroneous detection or erroneous judgment that occurs when performing the above.

For example, by including several pixels in both the left and right directions in several lines above and below the pixel that is the direct target of motion detection, the pixel that is the direct target of motion detection is In addition, the motion detection results for only that pixel and the motion detection results for surrounding pixel groups are prevented from differing, thereby preventing the occurrence of erroneous detection or erroneous judgment as described above. Furthermore, the presence or absence of movement is comprehensively determined after weighting is applied such that the pixel that is the direct target is the largest and the weight decreases as the distance from the pixel increases. According to such comprehensive determination of the presence or absence of motion, it is possible to obtain accurate switching signals by preventing erroneous determinations that may occur if the determination is made simply for each pixel.

FIG. 1 shows a configuration example of a motion compensation noise reduction device according to the present invention based on the above-mentioned operating principle. In the illustrated configuration, the image signal of the current frame from the input terminal is supplied to the subtracter 1, the difference between the predicted image signal formed as described later and the image signal of the current frame is calculated, and the subtracted output signal is After multiplying the signal by a coefficient not exceeding 1, the signal is supplied to an adder 3, and the predicted image signal and the current frame image signal are added together to form an output image signal, which is taken out from the output terminal, and the frame It is supplied to the memory 4 and stored therein in preparation for forming a predicted image signal. The image signal of the previous frame read out from the frame memory 4 with a delay of one frame period is
The motion detection circuit 5, the motion correction circuit 6, the switching judgment circuit 7, and the switching unit 8 are supplied in parallel, and as described later, based on the judgment of the switching judgment circuit 7, the motion-corrected predicted image signal and the non-motion-corrected predicted image signal are The signal is switched by the switch 8 and guided as a predicted image signal to the subtracter 1, and the difference between it and the current frame image signal from the input terminal is determined as described above. Furthermore, the image signal of the previous frame supplied to the motion detection circuit 5 is compared with the current frame image signal from the input terminal to detect the presence or absence of image movement, and the motion spectrum obtained as a result of the motion detection is subjected to motion correction. 1 supplied to the circuit 6 and supplied from the frame memory 4
A motion-corrected predicted image signal is formed by correcting the image motion of the image signal before the frame, and the motion-corrected predicted image signal is passed through a switch 8 under the judgment of a switching judgment circuit 7, which will be described later. The signal is led to subtractor 1 in the same manner as described above. The switching judgment circuit 7 receives the above-mentioned motion-corrected predicted image signal and the current frame image signal from the input terminal together with the above-mentioned non-motion-compensated predicted image signal, and generates both non-motion-compensated and motion-compensated predicted images. It is determined which of the signals closely approximates the current frame image signal, and based on the determination result, the predicted image signal that closely approximates the current frame image signal is selected via the switch 8 as described above. The signal is supplied to the subtracter 1 as well as the adder 3, and the weighted interframe difference signal from the coefficient unit 2 is added to the predicted image signal to form a desired noise-reduced output image signal. do.

Next, FIG. 2 shows an example of the configuration of the switching determination circuit 7, which is a main part of the motion compensation noise reduction apparatus of the present invention, which operates as described above with the configuration shown in FIG. Therefore, as described above, the switching judgment circuit 7 compares the non-motion compensated predicted image signal and the motion compensated predicted image signal with the current frame input image signal, and selects the predicted image signal that more closely approximates the current frame image signal. Subtractor 1 and adder 3 via switch 8
In order to supply the motion-corrected predicted image signal to the subtracter 1 and the adder 3, the switch 8 is controlled when the judgment output signal reaches a high logic level H. There is. Note that if the switching determination circuit 7 makes only such a determination, there is a risk that erroneous switching may occur due to noise, so as will be described later with reference to FIGS. The presence or absence of image movement in the pixel group that has been selected is comprehensively judged, and when the directly targeted pixel is in an isolated state, the switching judgment result for that pixel is discarded, and judgment is made by majority vote or weighted judgment. Then, a switching signal is output based on the final judgment result.

Therefore, in the switching determination circuit 7 having the configuration shown in FIG. 2, the non-motion compensated predicted image signal and the motion compensated predicted image signal from each input terminal are supplied to the subtracters 9 and 10, and are combined with the current frame image signal. Find the difference between the two, lead those difference signals to absolute value units 11 and 12 to find their respective absolute values, and supply the absolute values of these differences to amplitude comparators 15 and 16, respectively, until all of the following conditions are satisfied. The AND circuit 18, which is supplied with all the output signals of the amplitude comparators 15, 16, etc., determines that the motion-corrected predicted image signal should be used for noise reduction when The configuration is such that the AND output of is a high logic level H. That is, condition (1): When the difference between the non-motion compensated predicted image signal and the current frame image signal is greater than the level set by the switch register 13.

This condition (1) indicates that there is a movement in the image that causes a large difference between the predicted image signal and the current frame image signal unless motion correction is performed. In other words, since signal processing for motion correction has to be performed comprehensively on the pixel group of the image block, the still pixels move as the moving pixels are corrected, resulting in a new deviation from the current frame image signal. This is to prevent this, and when the difference is smaller than the level indicated by the switch register 13, it indicates that still pixels, which do not need to use a motion-corrected predicted image signal, are directly targeted for motion detection. .

Condition (2): When the difference between the motion-corrected predicted image signal and the current frame image signal is smaller than the level set by the switch register 14.

This condition (2) indicates that the motion correction is properly performed and the image signal predicted by applying the motion correction is more similar to the current frame image signal. In other words, the noise reduction device handles image signals with a lot of noise, and when there is a lot of noise, the noise mixed into the motion detection results strongly affects the motion correction results, making it difficult to perform proper motion correction. There is a possibility that the correction will not be performed accurately, and block-like distortion will appear in the image as a result of the motion correction. Therefore, on the other hand, if proper motion correction is performed accurately, the difference between the motion-corrected predicted image signal and the current frame image signal should not be that large, and if the difference becomes large, the motion This indicates that the correction was not performed properly and there was an error in motion correction, and that a non-motion corrected predicted image signal should be used for noise reduction.

Condition (3): When the amplitude comparator 17 compares the difference between the non-motion compensated predicted image signal and the motion compensated predicted image signal with the current frame image signal, and the latter difference is smaller.

This condition (3) literally indicates that the predicted image signal that is more similar to the current frame image signal is used for noise reduction.

In short, the above three conditions mean that between the motion-compensated predicted image signal and the non-motion-compensated predicted image signal, the one that more closely approximates the current frame image signal should be used for noise reduction, but which one is better? When determining whether the pixel is approximate, the condition is that if it is determined that the pixel is a still pixel or if it is determined that there was an error in motion compensation, the non-motion compensated predicted image signal is used.

Therefore, when determining whether to switch between a motion-compensated predicted image signal and a non-motion-compensated predicted image signal as described above, if only a single pixel is directly considered, errors will occur in the judgment due to the influence of noise. As mentioned above, this is a possibility. Therefore, FIGS. 3 and 4 respectively show examples of the configuration of a switching judgment device that can make a judgment about a pixel that is a direct object of judgment while referring to the state of a group of pixels surrounding the pixel that is a direct object of judgment. show.

The switching determination circuit having the configuration shown in FIG. 3 makes a determination by referring only to the pixel group on the same horizontal scanning line as the directly targeted pixel. _{-1 to -3} to sequentially delay each pixel period, and sequentially supply the image signals shifted by one pixel to the isolated point removal circuit 20 and compare the sequential pixel signals with each other, A pixel signal at an isolated point that is different from the previous and subsequent pixel signals is removed. That is, in the isolated point removal circuit 20, for example, for three sequential pixels, when both the preceding and succeeding pixel signals are at a high logic level H, the directly targeted pixel signal is at a low logic level L, or vice versa. In some cases, the directly targeted pixel signal is removed as an isolated signal, that is, a signal that is inappropriate to be used as a determination target due to the influence of noise or the like. Note that this level comparison can be similarly performed for a plurality of front and rear pixels by increasing the number of cascade-connected one-pixel delay devices 19. The image signal from which the isolated points have been removed as described above is supplied to the majority judgment circuit 22 directly and sequentially through the one-pixel delay devices 21-1 _{to 21-4} , and the one-pixel A decision is made by majority vote regarding the image signals shifted by a minute. That is, in the illustrated example, for five sequential image signal groups, if three or more pixel signals are at a high logic level H, the majority decision output signal is also set to a high logic level H, or if three or more pixel signals If the signal is at a low logic level L, the majority decision output signal is also set to a low logic level L, and the aforementioned switching of the predicted image signal is controlled by the majority decision output signal.

In the switching judgment circuit having the above-mentioned configuration, one-dimensional judgment was made by focusing only on the pixel group horizontally adjacent to the directly targeted pixel, but in order to make more appropriate and accurate judgment, two The configuration example shown in Figure 4 is designed to make judgments about dimensional pixel groups, and the image signal from the input terminal is processed by inserting a one-line memory 25 _-1,2 in the middle to determine the target horizontal scanning line. A plurality of 1-pixel delay devices 24 _{-1 to 10} are installed for each scanning line before and after the 1-pixel delay device 23 that takes out the target pixel signal directly while sequentially moving the signal.
are connected in cascade, and a total of 11 pixel signal groups of 3 pixels on the top and bottom and 2 pixels on the left and right taken out from these 1-pixel delay devices are supplied to the judgment device groups 26-1 _{to 26-4} to perform the above-mentioned two-dimensional judgment. It's summery. In addition, each 1 line memory
25 _-1,2 is designed to give the image signal a delay equal to, strictly speaking, one line minus four pixels. Further, the decision device groups 26-1 _{to 26-4} perform decisions including isolated point removal, majority decision, and weighting similar to those in the configuration example shown in FIG. Output a signal. If the decision algorithm is relatively simple, such a decision device can be constructed using only logic circuits; however, if the decision algorithm is complex, a read-on memory ROM may be used to By storing judgment results for various fixed patterns in advance,
This can be realized without complicating the circuit configuration.

Furthermore, as mentioned above, the reason why a pixel that is directly targeted is judged by referring to the surrounding pixel group is to prevent the occurrence of erroneous judgments based on noise in the input image signal. If the ratio is good, there is no need to refer to surrounding pixel groups; on the contrary, if judgments are made including unnecessary pixel groups, it will not be possible to make accurate judgments on the target pixel directly. The opposite effect also occurs. Therefore, in the configuration shown in FIG. 4, a plurality of judgment devices are provided in which the judgment algorithm and the range of the surrounding pixel group to be referred to are sequentially different. Switcher 27
can be operated manually or automatically to make decisions at appropriate stages.

As is clear from the above description, according to the present invention, a predicted image signal obtained by applying motion correction to a one-frame delayed image signal used for noise reduction using interframe correlation of image signals, that is, a motion-corrected Predicted image signal and simply 1 without motion correction
The frame-delayed predicted image signal, i.e.
The non-motion-corrected predicted image signal is switched based on the result of determining the presence or absence of image movement for each pixel, and the predicted image signal that approximates the current frame image signal is used for noise reduction. , to prevent disadvantages such as images of moving parts becoming blurred due to afterimages caused by noise reduction, and preventing such blurring from causing noise in moving parts, and to prevent imperfections in motion correction for parts of intense movement. Therefore, in order to prevent new image quality deterioration such as block distortion in moving parts and accompanying movement of still pixels that occur when noise reduction is performed using a motion-corrected predicted image signal, the above-mentioned method is used. When making decisions for each pixel, refer to the state of surrounding pixel groups,
Based on the result of the comprehensive judgment, it is possible to switch and use predicted image signals that are much more accurate than in the past, and to perform extremely appropriate noise reduction using interframe correlation of image signals.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an example of the configuration of a motion compensation noise reduction device of the present invention, and FIG. 2 is a block diagram showing an example of the configuration of a switching judge used in the device of the present invention.
FIGS. 3 and 4 are block diagrams respectively showing other configuration examples of the switching determination device. 1, 9, 10...Subtractor, 2...Coefficient unit, 3...Adder, 4...Frame memory, 5...Motion detection circuit, 6
...Motion correction circuit, 7...Switching judgment circuit, 8, 27
...Switcher, 11, 12...Absolute value device, 13, 14
...Switch register, 15, 16, 17... Amplitude comparator, 18... AND circuit, 19 _{-1 to 3} , 21 _{-1 to 4} ,
23, _{24 -1 to 10} ...1 pixel delay device, 20...isolated point removal circuit, 22...majority decision circuit 22, 25 _-1,2 ...
1 line memory, 26 _{-1 to 4} ...Judgment device.

Claims (1)

[Scope of Claims] 1. The input image signal is determined by the interframe correlation between the input image signal and a predicted image signal obtained by correcting the position of the image according to the moving direction and size of the image with respect to the input image signal one frame before. A motion compensation noise reduction device for reducing noise in the input image signal, for each successive pixel in the input image signal, detecting image motion for a plurality of pixels including pixels adjacent to each pixel, and detecting the motion of the image. Based on the results of comparing a motion-corrected predicted image signal subjected to position correction according to direction and magnitude and a non-motion-corrected predicted image signal without position correction with the input image signal to determine their respective approximations, A motion-compensated noise reduction device characterized in that the motion-compensated predicted image signal and the non-motion-compensated predicted image signal are switched for each successive pixel to produce the predicted image signal. 2. In the motion-compensated noise reduction device according to claim 1, of the motion-compensated predicted image signal and the non-motion-compensated predicted image signal, the one having a smaller difference from the input image signal is used as the non-motion-compensated predicted image signal. The difference between the corrected predicted image signal and the input image signal is larger than a predetermined value, and
A motion-compensated noise reduction device characterized in that the motion-compensated predicted image signal is set as the predicted image signal only when a difference between the motion-compensated predicted image signal and the input image signal is smaller than a predetermined value.