KR960013234B1 - Koga - 3 stop motion vector detecting method for movement detecting apparatus - Google Patents

Abstract

The method relates to carrying out serve sampling for image data of the image data block to detect a similar image data block. The detection method comprises the steps of: detecting image data block within the similar frame memory to the input image data block of the input image memory; transferring a candidate vector to be the second step among the routine 1; and changing the candidate vector to be the third step among the routine 1.

Description

Nose-3 motion vector detection method of motion detector

1 is a block diagram of a typical image compression transmission device.

2 is another block diagram of a typical image compression transmission device.

3 is a block diagram of a motion detection device constructed in a general image compression transmission device.

4 (a) (b) and (c) are diagrams showing states in which image data blocks move.

Figure 5 is a state diagram showing a conventional koga-3 step motion vector detection method.

6 is a flowchart illustrating a method for detecting a motion of a No. 3 step in a motion detection apparatus according to the present invention.

Figure 7 (a) (b) is a state diagram showing a sub-sampling method performed in the Koga-3 step motion vector detection method of the motion detection apparatus according to the present invention.

* Explanation of symbols for main parts of the drawings

1: DCT unit 2: quantization unit

3: variable length encoder 4: frame memory

5: subtractor 6: inverse quantization unit

7: reverse DCT section 8: adder

9: motion detection device 10: input image memory

11: ROM 12: CPU (Central Processing Unit)

The present invention relates to motion detection. More particularly, the present invention relates to a method for detecting a Koga-3 step motion vector of a motion detection apparatus suitable for easily detecting similar blocks among input image block data and image data stored in a memory. will be.

With the progress of the information society, there is a strong demand for the accumulation and transmission of natural images in addition to the accumulation and transmission of text, numerical values, and voice data.

In the case of accumulating or transmitting such natural images, in order to accumulate a large number of images or to perform high-speed transfer, it is necessary to greatly compress the data amount of a huge image by image input processing.

Currently, the most attention-grabbing image compression process is a discrete cosine transform (hereinafter referred to as DCT) coding.

Since the DCT coding method can greatly compress the amount of data, and the quality of the reproduced picture is also good, it is decided to adopt it as the basic method of the internationalized video coding standard that has been examined by ISO / CCITT.

In the DCT coding method, first, the input format is divided into square blocks about 8 by 8 pixels and applied to the DCT unit 1 as shown in FIG. 1, and the DCT unit 1 performs two-dimensional DCT conversion for each block. Will be performed.

The DCT conversion is a process of converting an image from a spatial domain to a frequency domain. In general, since an image is composed of low frequency components, a significant data compression of the image is possible by encoding only the low frequency components of the DCT coefficient after conversion.

In order to encode the DCT coefficients into small data amounts, the quantization unit 2 performs quantization processing of the DCT coefficients. Then, the variable length coding unit 3 performs variable length coding on the quantization index indicating the DCT coefficient value after quantization using a variable length code such as a Hoffman code to output the final code.

At this time, an image data block similar to the input image data block input to the DCT unit 1 is detected among the image data that has already been transmitted, the DCT is changed for the difference value, and then encoded and transmitted. By adding the difference values for the blocks, it is possible to obtain the same image data as the input image data.

Accordingly, another conventional image compression transmission device allows the image data corresponding to all frames of the input image data block to be stored in the frame memory 4 as shown in FIG.

Then, the motion detection device 9 detects in the frame memory 4 an image data block similar to the input image data block based on the telephone image data of the frame memory 4.

At this time, if a similar image data block is found, the motion detection device 9 transmits the positional relationship in which the input image data block and the similar image data block are located in the frame to the receiver side as a motion vector, while the subtractor 5 Subtracts the image data for the similar image data block of the frame memory 4 from the input image data.

Then, the input image data will be further compressed by the DCT conversion and quantization of the difference value of the image data added by the subtractor 5. At this time, the image data quantized by the quantization unit 2 is decoded by the inverse quantum unit 6 and the inverse DCT unit 7, and then the likeness of the similar image data block output by the frame memory 4 is the same as the receiving unit. By adding the image data and the adder 8, the same image data as the input image data is obtained.

Since the image data of the adder 8 is stored at the same position on the frame by the motion vector by the motion detection device 9, the frame memory 4 always has an image corresponding to all frames of the input image data. To store data.

Therefore, the compression rate of the data by the DCT is different depending on how the motion detection device 9 detects in the frame memory 4 an image data block similar to the input image data block.

3 is a block diagram of a conventional motion detection apparatus for implementing the KOGA-3 step motion vector detection method for detecting a motion vector.

The frame memory 4 is the same as the frame memory of FIG. 2, and the input image memory 10 inputs an input image data block to indicate which position in the frame corresponds to an internal central processing unit (hereinafter referred to as CPU) (with an address signal) ( 12).

At this time, the CPU 12 outputs and calculates a candidate vector for executing the step 3, which will be described later.

The ROM 11 is configured to store candidate vectors as address values and apply them to the frame memory 4 for address calculation.

In the first step of the Coca-3 step motion vector detection method, when the corresponding address of the input image data block is applied to the CPU 12 to the input image memory 10 in the above-described motion detection apparatus, the CPU 12 is a candidate. The address storing the vector is applied to the ROM 11 to check the image data in the moving compensation area.

Specifically, the block shown in Fig. 4A is assumed to be an input image data block. The CPU 12 applies an address corresponding to the position of the input image data block within the frame to the input image memory 10.

The CPU 12 then applies to the ROM 11 an address specifying a candidate vector to designate a block of image data in the frame memory 4 at the same position as the address applied to the input image memory 10.

The ROM 11 outputs a candidate vector, and applies an address in the same area as the input image data block in the frame memory 4 so that the image data block for the address from the frame memory 4 is transferred to the CPU 12. Is authorized.

In other words, the CPU 12 receives the image data block shown in Fig. 4B. In this way, the image data in the frame memory 4 having the same address as the input image data block is much smaller than the case where the image of the previous frame and the current frame changes abruptly. This is because the image data in the frame memory 4 is most likely to be most similar.

The CPU 12 then determines the similarity by subtracting the sum of the image data of the input image data block and the sum of the image data block in the frame memory at the same address.

Then, the CPU 12 changes the position vector and sequentially checks the input image data block and the image data block in the adjacent frame memory 4 to detect the image data block most similar to the input image data.

Referring to FIG. 5, the block A is an image data block in the frame memory 4 at the same address as the input image data block as described above.

In this case, the center point A1 of the block is called a search point. As described above, the CPU 12, which has determined the similarity degree between the image data block A having the search point and the input image data, changes the position vector so that the image data block of the address different from the block A is framed. It is applied from the memory 4.

At this time, as shown in Fig. 5, the degree of different address of the image data block is moved to the search points A1 up, down, left, and right so as to have four pixel intervals, and thus, the search points A3, A8, A5, A6 and A6. Is moved to form a position vector to form search points A2, A7, A4, and A9 up and down four pixel intervals from the search points A5 and A6.

The CPU 12 detects the image data block most similar to the input image data block among the image data blocks formed from the search points A1 to A9.

At this time, the determination of the similarity is performed by the search points A1 to A9 from the sum of the respective data of the input image data blocks, in the same manner as the method of determining whether the block A is similar to the input image data block. The block forming the smallest difference value by subtracting the sum of the respective data is determined as the most similar block.

In the case of this embodiment, it is assumed that the block B formed by the search point A7 is determined to be most similar to the input image data block.

The CPU 12 proceeds to the second step of the Koga-3 step motion vector detection method, moves the search point again from the search point A7, and a new image data block transfers the image data to the frame memory 4. Will be input from

That is, the CPU 12 moves the search points B2, B7, B4, and B5 up, down, left, and right, two pixels apart from the search point A7, and the pixels from the search points B4 and B5. The position vector is changed to form two search points B1, B2, B3, and B8 separated up and down, and the image by the search points B1 to B8 in accordance with the address signal of the ROM 11 is changed. The data block is applied to the CPU 12.

The CPU 12 inputs the image data of the image data block by the new search points B1 to B8 and the image data of the image data block by the two-level search points centered on the search point A7 according to the first step. What is determined to be the most similar to the image data block is detected by performing the same method as the image data block by the search points A1 to A9.

At this time, it is assumed that the image data block C by the search point B6 is determined to be most similar to the input image data block.

If it is determined that the image data block by the search point B6 is the most similar, the CPU 12 proceeds to the third step of the Koga-3 step and changes the position vector again, and the pixel from the search point B6 is changed. To form search points C2, C6, C4, and C6 separated by one from each other, and search points C1, C6, C3, and C8 separated by one pixel up and down from the search points C4 and C5. Change the position vector.

As described above, the input image data block of the image data block of the search points C1 to C8 and the image data block of the three step search points centered on the search point B6 of the second step; The most similar image data block is detected.

In this example, the image data block D by the search point C6 is shown to be the most similar. That is, the Koga-3 step motion detection method detects an image data block most similar to the input image data block among the image data blocks within a predetermined distance from the input image data block and the image data block of the same address in the frame memory 4, The movement relationship is output as a motion vector.

At this time, the method in the KOGA-3 step changes the search point in the order of 4-2-1 in the pixel interval from the image data block of the same address and the image data block of the same address, and replaces the similar image data block. It can be seen that the detection.

That is, in the three-stage motion vector detection method, as shown in Fig. 4B, image data in a predetermined motion compensation area E is obtained from a frame memory 4 image data block having the same address as the input image data block. Block to detect an image data block similar to the input image data block.

In this case, when the input image data is located at the edge of the screen or the like as shown in FIG. 4C, similar image data is detected in the periphery of the image data at the same address as the input image data. It will be more limited than the above-described processes.

However, the method of detecting the Koga-3 step motion vector described above is the most similar image data as the difference between the sum of the total image data of the input image data block and the sum of the total image data of the predetermined block in the frame memory. The block is detected so that the computational power of the CPU must be high.

That is, since a general image data block is composed of 16 x 16 pixels, the motion of the CPU is caused by the operation of adding all the image data of the image data block and the similar image data block and calculating the difference between the added values and the like. There was a problem that the detection device can not be compact, high-speed computerized.

SUMMARY OF THE INVENTION The present invention has been made to solve such a problem, and an object of the present invention is to provide a motion detection apparatus capable of miniaturizing and high-speed computerization of a motion detection apparatus by subsampling image data of an image data block and detecting a paid image data block. Noga-3 step motion vector detection method is provided.

According to the present invention for achieving this object, an image of the input image memory 10 using the motion detection device comprising the input image memory 10 CPU 12 and the ROM 11 and the frame memory 4 is used. The image data blocks in the frame memory 4, which are similar to the input image data blocks, are detected by comparing the image data of the image data blocks in the frame memory 4 formed by moving the image data constituting the data block and the search point in three steps. A nosy-3 step motion detection method of a motion detection apparatus, comprising: inputting an address of input image block data from the input image memory 10 (S1), and outputting a candidate vector according to the input address; (S2) and receiving as image data of an image data block of the frame memory 4 corresponding to an address designated by the ROM 11 in accordance with the candidate vector. (S3), the step S4 of subsampling the input image data, the total sum of the sampled input image data, and the sampled image data among the image block data blocks input from the frame memory 4; It is determined whether the step (S5) for obtaining the difference value of the step, the step (S6) for storing the difference value, and whether the steps (S1 to S6) were performed a total of nine times. S6) is repeated nine times, and if it has been performed nine times, a step S8 of detecting a similar block by selecting the smallest value among the stored nine difference values, and is related to the detected similar block. And outputting a final motion vector obtained by moving and changing the candidate vector.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 6 is a flowchart for implementing a method of detecting a Koga-3 step motion vector of a motion detection apparatus according to the present invention, and includes 1, 2, and 3 routines L1, L2, and L3.

Specifically, the routine L1 is used to detect an image data block in the frame memory 4 similar to the input image data block of the input image memory 10 in accordance with the step of Coga-3. S8).

In this case, step S1 is a process of inputting an address of input image block data from the input image memory 10 to the CPU 12, and step S2 is a process for outputting a candidate vector according to the input address. to be.

Step S3 is a process for inputting the image data of the image data block of the frame memory 4 according to the address designated by the ROM 11 according to the candidate vector, and step S4 is the input image data. This is a process for subsampling them.

At this time, the subsampling in step S4 samples the image data of even or odd rows among the image data constituting the image block data as shown in Fig. 7A.

Step S5 is a process of calculating the total value of the sampled input image data and the difference value of the sampled image data among the image block data blocks input from the frame memory 4, and step S6 stores the difference value obtained. It's a process.

In this case, step S7 is a process of determining whether the search points in the first step of the Koga-3 step motion vector detection method are moved nine times by determining whether the steps S1 to S6 have been performed a total of nine times. S8) is a process of selecting the smallest value among the nine difference values stored.

That is, step S8 is a process of detecting the image data block most similar to the input image data in the first step of the Koga-3 step vector detection method.

At this time, the candidate vector in the step (S2) is moved in the same manner as the first step of detecting the Koga-3 motion vectors. The routine L2 and L3 perform the same process as the routine L1, but the routine L2 is a candidate to correspond to step 2 of the Coca-3 step motion vector detection method in the step S2. The vector is moved, and the routine L3 changes the candidate vector so as to correspond to the third step of the nose-3 motion vector detection method.

At this time, the subsampling method in the routine L2 is the same as the routine L1, but the subsampling method in the routine L3 causes the image data to be sampled across one row for every row as shown in FIG. Is done. In addition, since the address of the input image data block is already applied to the CPU 12 in step S1, it is unnecessary in the routines L2 and L3.

Further, in step 7, when located at the corner of the screen or the like of the image data block, the number of movements of the search point possible in the corresponding movement compensation area will be lowered.

As described above, it can be seen that the method of detecting the KOGA-3 motion vector of the motion detecting apparatus according to the present invention configured as described above is performed in the same order as the conventional KOGA-3 motion vector detecting method.

However, in the conventional KOGA-3 step method, the similarity degree between the input image data block and the image data block of the frame memory 4 is obtained by adding up all the image data of the input image data block and the image data block of the frame memory 4. The image data is determined according to the difference of the sum of all the image data. However, in the present invention, the image data is determined as the difference with respect to the subsampled value.

That is, in the present invention, in step 1 and 2 of the method of detecting the nose-3 step, the image data constituting the image data of the input image data block and the image data block of the frame memory 4 is shown in FIG. Subsampling Then, the difference between the sum of the image data of the subsampled input image data block and the image data sum of the frame memory is determined to determine whether the similarity exists.

In general, considering the fact that an image has more non-moving images than a moving image, even if it is subsampled as shown in FIG. It can be seen.

In addition, in the present invention, by subsampling as shown in Fig. 7 (b) in step 3 of the koga-3 step, it is possible to further improve the certainty of the similarity determination. That is, when steps 1 and 2 of the Koga-3 step motion vector detection method are performed, the image data blocks around the candidate search points according to the step 2 may have similar image data blocks.

Therefore, in the third step of the Koga-3 step motion vector detection method, the most similar image data block is obtained by subsampling the image data of the image data block one column across all rows as shown in FIG. Can be detected.

As described above, the present invention detects a similar image data block by subsampling the image data of the input image memory and the image data of the frame memory with the subsampling value, thereby reducing the amount of computation of the CPU and reducing the computational capacity of the CPU. Since it can be designed small, the motion detection device can be miniaturized and high speed.

Claims (2)

Input image memory 10 The motion detection device comprising the CPU 12 and the ROM 11, and the image data and the search point constituting the image data block of the input image memory 10 by using the frame memory 4 The method of detecting the motion of the motion detection apparatus of the motion detection apparatus for detecting the image data blocks in the frame memory 4 similar to the input image data blow by comparing the image data of the image data blocks in the frame memory 4 formed by moving to the step. In step S1 of inputting an address of input image block data from the input image memory 10, a step S2 of outputting a candidate vector according to the input address, and the ROM according to the candidate vector. Step S3, which is input as image data of the image data block of the frame memory 4 according to the address designated by 11, and subsampling the input image data. Is a step S4, a step S5 for calculating a difference value between the total of the sampled input image data, the sampled image data among the image block data blocks input from the frame memory 4, and the difference value. As a result of determining whether the step S6 to be stored and the steps S1 to S6 have been performed a total of nine times, if not performed nine times, is repeated until the steps S2 to S6 have been performed nine times, and Performed nine times, the step S8 of detecting a similar block by selecting the smallest value among the stored nine difference values, and outputs a final motion vector obtained by moving and changing a rear vector related to the detected similar block. Noga-3 step motion vector detection method of a motion detection device characterized in that. The method of claim 1, wherein the sub-sampling is performed by sub-sampling only image data of even or odd rows in the first and second steps of the Koga-3 motion vector detection method. A method for detecting a motion vector of a motion detector of a motion detection apparatus, characterized in that sub-sampling is performed for each column.