JPH0671334B2 - Accuracy determination method of motion amount estimation value of moving image - Google Patents

Description

TECHNICAL FIELD The present invention relates to a method for determining the accuracy of an estimated value of a motion amount in a moving image.

(Prior Art) As one of the moving image signal coding methods, research and development have been actively conducted on a high-efficiency interframe coding method using a motion compensation technology and a standard frame number conversion method. In order to perform this motion compensation, a technique for estimating the motion amount of a moving image signal with high accuracy is required, and there is a method of performing it for each sampled pixel signal (hereinafter referred to as “pixel”) and a certain method. There is a method of estimating a motion amount by grouping several pixels into a block. Of these, the latter method of estimating the motion amount in block units is generally widely adopted, and therefore this method will be described in detail in the following description.
The former method of performing pixel units can be performed in substantially the same manner by replacing the latter block unit with pixel units.

When estimating the amount of motion of a moving image signal in block units, one frame is divided into block units consisting of, for example, 8 pixels in the horizontal direction and 8 lines in the vertical direction, for a total of 64 pixels. Next, for each block unit in the current frame, from the immediately preceding frame, (1) A method of estimating the amount of motion using the similarity of the signal pattern in the block (JP-A-55-162683-5).

(2) A method for estimating the amount of motion by physically correlating the in-frame signal gradient and the inter-frame signal difference value (Japanese Patent Laid-Open No. Sho 60)
-158786).

Etc. are known.

(Problems to be Solved by the Invention) Each of the three types of motion amount detection methods described above uses the similarity of intra-block signals between frames in some form, and is highly accurate for relatively small motions. An estimate of is obtained. However, it is difficult to accurately estimate the amount of motion when an object moves at high speed in a stationary background.

FIG. 3 is an example of a conventional technique, and is an explanatory diagram when an object moves at high speed in a still background. FIG. 3A is a screen of the current field, FIG. 3B is a screen one frame time before, and FIG. (C)
Shows the screens one field before. In the above-described motion amount detection method, the similarity between the screen (b) which is the screen one frame before and the screen (a) which is the screen of the current field is calculated. An object 3 having almost the same width as that of the estimation block 2 is quickly shown in FIG.
Therefore, if it moves as shown in FIG. 3A, the movement of the object 3 is not detected at all in the block 2. Therefore, if it is necessary to interpolate a field between the current field and the field one frame before as shown in FIG. 7C by the frame number conversion (NTSC → PAL or SECAM), block 2
The passage of the object 3 at the portion corresponding to is not reflected at all, and the movement of the object 2 becomes unnatural.

Therefore, a technology that detects whether or not the estimated value of the motion amount is almost accurate, detects the accuracy of the estimated value, uses the estimated value as it is when the accuracy is good, and improves by field interlacing when the accuracy is poor. is necessary. However, until now, there was no technical idea to judge the accuracy of the estimated value of the motion amount.

The present invention has been made in order to solve the above-mentioned problems of the conventional technique, and an object thereof is to verify the accuracy of the estimated value of the motion amount in a moving image and determine whether the estimated value is correct.

(Means for Solving Problems) A feature of the present invention is to obtain an inter-frame difference and an inter-field difference corrected by the estimated motion amount, and determine that the sum of absolute differences between the inter-frame difference and the inter-field difference is maximum. Is used as the accuracy result of the estimated amount of motion.

(Principle of the Invention) As the following moving image, a television signal will be described as an example.

Even-numbered fields and odd-numbered fields are alternately transmitted in the television signal, and in the estimation of the motion amount, the estimation processing is performed between frames, that is, between two fields skipped. Therefore, for example, as shown in FIGS. 3A and 3B, no movement of the object is detected in the block 2. In the field existing between the two fields used for these estimations, an object moving to the position of block 2 appears as shown in FIG. 3 (c). That is, even if the motion amount is estimated based on the pattern similarity between the frames when the motion amount is estimated between the two fields separated by one frame time, the pattern is corrected between the fields by the estimated motion amount. If no similarity is found, it can be determined that the accuracy of the estimated value is low.

On the contrary, if the similarity between the patterns corrected by the motion amount estimated between the fields separated by 1 field time and 1 frame time is high, the estimation accuracy can be determined to be high. Such pattern similarity can be evaluated by, for example, an absolute difference between fields of the corresponding block. This is expressed by the following equation.

: Current field : Previous field (after motion compensation) : 1 field before frame (after motion compensation) C: motion estimation block m × m: block size i, j: motion estimation block address (i, j ∈ c) P: pattern similarity The verification of the motion estimation value can be performed by calculating the following parameter α.

That is, the first term in the equation (2) is the block one field before, which is corrected by the motion amount estimated from the block Ai, j with the current field. The difference between the fields is calculated by subtracting
The term is the block one frame before that corrected by the amount of motion estimated from the block Ai, j with the current field The inter-frame difference is obtained by subtracting, and the maximum value of the absolute difference sum of the inter-field difference and the absolute difference sum of the inter-frame difference is defined as the motion amount estimated value accuracy in the present invention.

(Example) Hereinafter, the present invention will be described in detail with reference to the drawings. In addition,
In the following description, it is assumed that the estimated value of the motion amount has already been obtained by the above-mentioned conventional method, and only the method of determining the estimated value accuracy will be described.

FIG. 1 is an embodiment of the present invention and is a block diagram of a method for determining an estimated value accuracy of a motion amount. The motion amount estimated value signal 4 is input to the address control unit 5, and the field memory address signals 6, 7 and 8 corrected by the motion amount estimated value are generated. On the other hand, the image signal 9 is a field memory (hereinafter,
"FM"), the current field is FM10, 1
The image signal before the field is stored in FM11 and is stored in the image signal FM12 before one frame. The image signal stored in each FM 10, 11 and 12 is taken out as a signal (the amount of motion is corrected) which is deviated by the estimated value of the amount of motion by each address signal 6, 7 and 8 and the corrected one field before The image signal 14 and the image signal 13 of the current field are input to the difference calculation circuit 16, and the absolute value of the inter-field difference corrected by the motion amount is extracted. Similarly, the image signal 15 of one frame before and the image signal 13 of the current field, which are similarly corrected, are extracted by the difference calculation circuit 17 to obtain the absolute value of the inter-field difference corrected by the amount of motion. To enter. The maximum value detection circuit 18 compares the absolute value of the inter-field difference with the absolute value of the inter-frame difference to obtain the maximum value. This maximum value becomes the estimated value precision parameter α of the motion amount to be obtained. Further, the estimated value accuracy parameter α is input to the determination circuit 19 and compared with a predetermined threshold value T, and when the estimated value accuracy parameter α is smaller than the threshold value T, the estimated value of the motion amount is correct (usable). To determine. Conversely, when the value of the estimated value accuracy parameter α is larger than the threshold value,
It is determined that the estimated value of the obtained motion amount is bad (unusable). Here, when the determined result is unusable, for example, the estimated value accuracy parameter α may be used as the estimated value of the motion amount, or the initial deviation filed on the same day by the same inventor. The amount of movement may be corrected by another method, or the amount of movement may be obtained again by another method.

Next, regarding the value of the threshold T, it varies depending on the moving image. Second
The figure shows the experimental results obtained by simulating the results of the conventional inter-frame motion amount estimation method for a normal television signal using the estimation value accuracy determination method of the present invention. The figure shows the standard amplitude of the television signal is 1.0V.
(Of which, the video part is 0.7V) is quantized with 512 levels of 9 bits and digitally processed. 0.7 × 512 level of the digital signal is the maximum amplitude of the video. FIG. 6 is a diagram in which the horizontal axis represents the estimated value accuracy parameter α and the vertical axis represents the occurrence frequency of the estimated value accuracy parameter α. It can be seen that the estimated amount of motion significantly deteriorates when the estimated value accuracy parameter α becomes greater than about 10. Therefore, in the case of a normal television signal, the threshold value T may be set to about 10. In the above description, a normal television signal is described as an example, but the present invention is not limited to this, and can be applied to all moving images using the inter-frame motion amount estimation method.

(Effect of the Invention) As described in detail above, according to the present invention, the estimated value accuracy parameter of the motion amount is obtained using the absolute value of the inter-frame difference and the absolute value of the inter-field difference in which the motion amount is corrected, and the parameter is calculated. It is possible to prevent the use of an incorrect estimation value when a highly accurate estimation value of a motion amount such as a moving image frame number conversion method is required and always obtain a high quality image signal. Is possible and its effect is extremely large.

[Brief description of drawings]

FIG. 1 shows an embodiment of the present invention, and is a block diagram of motion amount estimation value accuracy determination, and FIG. 2 shows an experimental result using the estimation value accuracy parameter of the present invention. FIG. 3 is an explanatory diagram for explaining an estimated value when an object moves at high speed by the conventional motion amount estimated value method. 1 ... Still background, 2 ... Motion amount estimation block, 3 ... Moving object, 4 ... Motion amount estimated value signal, 5 ... Address control unit, 6, 7, 8 ... Address signal, 9 ... Image signal, 10,1
1,12 ...... Feed memory, 13,14,15 …… Image signal with corrected motion amount, 16,17 …… Differential calculation circuit, 18 …… Maximum value detection circuit, 19 …… Judgment circuit.

Claims (1)

[Claims] 1. A motion amount which divides an image signal into blocks of a predetermined size, and estimates a motion amount between a current field and a previous field which is a field one frame before the current field for each block. In the estimated value method, the sum of values obtained by correcting the motion amount of the previous field, which is one field before the current field, by the estimated value and adding the absolute difference between the current field and the current field in the block. Is a first sum of absolute differences, and the sum of sums of absolute differences of inter-frame differences between the previous field and the current field corrected by the estimated value in the block is calculated as a second difference absolute value. As the sum of values, the first sum of absolute differences and the second sum of absolute differences are calculated for all blocks in each field, and the maximum absolute value of the difference is calculated. A sum is defined as an estimated value accuracy parameter of the motion amount, and when the estimated value accuracy parameter is smaller than a predetermined value, it is determined that the estimated value accuracy of the motion amount is good. Accuracy determination method.