JPS63181585A - Motion compensated interframe coding device for TV signals - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to high efficiency encoding of television (TV) signals, and more particularly to miniaturization and economicalization of motion compensated interframe prediction circuits.

[Conventional technology]

Since TV signals have a wide frequency band of 4 MHz or more, in order to directly convert them into digital signals and transmit them, a high transmission speed of about 100 Mbps is required. "Motion compensated interframe prediction" is known as a typical method for reducing this speed to 1.5 Mbps or less.

This method utilizes the correlation between consecutive screens (frames). That is, the current frame to be encoded T
Instead of transmitting the V signal as it is, do the following.

1) Error between the TV signal of the frame to be encoded and the TV signal of the previous frame, such as when the image is stationary (
When the difference between frames) is small, there is no difference between frames.
Only the information to that effect will be transmitted.

2) When the image moves and the difference between frames becomes large, detect the amount of image movement (direction and size), predict the current frame from the previous frame at the position that compensates for it, and calculate the prediction 8t.
! Encode the I error.

Hereinafter, the amount of motion of this image will be referred to as a motion vector.

It is inefficient to transmit the presence or absence of the above-described interframe difference and the motion vector for each TV signal. Fortunately, however, there is a strong correlation between the presence or absence of interframe differences and motion vectors between adjacent signals. So actually.

Multiple signals are combined into a block, and the presence or absence of interframe differences and motion vectors are detected and transmitted in block units.

There are two known motion vector detection methods: first order.

1) Exhaustive search method: Generate all motion vectors within the range to be searched, measure the similarity between the block in the previous frame and the block in the current frame at the corresponding position, and find the one with the highest degree of similarity. Select a motion vector.

2) Multi-step depth method: First, a small number of motion vectors are generated with coarse density, and the similarity of each is measured and compared. Next, it is assumed that there is an optimal motion vector around the motion vector with the highest degree of similarity among them, and motion vectors around it are generated with high density, and among them, the one with the highest degree of similarity is selected. Thereafter, the generation density of the motion vectors to be generated is increased as necessary, and the above operation is repeated.

[Problem that the invention seeks to solve]

The full search method has the advantage of being able to detect the motion vector with the highest degree of similarity within its search range and minimizing the prediction error between frames after motion compensation. However, since the number of searches is large, it is necessary to speed up the processing and parallelize the processing, resulting in a problem that one circuit becomes large-scale.

The multi-step search method also requires a predetermined number of searches. Therefore, when the number of encoding bits is extremely short, for example when the difference between frames is small and the fact that there is no significant difference between frames is encoded with just one bit, the period corresponding to the number of encoding bits is It is necessary to complete the above search. For this reason, although it is reduced compared to the exhaustive search method, it still requires high-speed processing and parallelization, and there is a problem that the circuit becomes large-scale.

There is also the problem that, due to the circuit constraints mentioned above, it is advisable to search a wider range.

[Means for solving problems]

The above problem is solved by making the number of searches approximately proportional to the number of encoded bits. That is, the search can be performed according to the following procedure.

1) Measure the similarity between a block in the current frame and a signal in the previous frame at the same position (referred to as motion vector 0). If the difference is greater than or equal to a predetermined value, it is immediately determined that there is no significant difference, a message to that effect is transmitted, and the search is terminated.

2) In other cases, similarly to the above-mentioned known multi-stage search method, motion vectors are generated at a coarse density, and those with high similarity, including motion vectors O, are selected.

Motion vector 0. When the motion vector or its vicinity is selected, a relatively small motion vector can be estimated to be optimal. Here, the following two points are generally known.

0 The frequency of occurrence of an optimal motion vector is higher as the amount of motion is smaller. 0 Therefore, if this is entropy encoded, the number of encoded bits of the motion vector will be shortened, and the time allowed for processing will also be shortened.

Therefore, in the present invention, in this case, only the motion vector O or a narrow range in the vicinity thereof is searched to reduce the number of searches.

When a motion vector other than 0 or its vicinity is selected, the optimal motion vector can be estimated to be a relatively large amount of motion. In this case, the code word length of the motion vector becomes longer, so the number of searches can be increased.

Therefore, a wide range is searched to detect motion vectors with higher similarity.

[Effect]

According to the above means, the optimum search means is selected depending on the magnitude of the amount of movement of the image, as shown below.

For a small amount of motion with a small number of 0-encoded bits,
Since the number of searches is also reduced, malfunctions due to insufficient processing will not occur.

Furthermore, since the number of searches can be increased for a large amount of motion with a large number of encoded bits, a wider range can be searched.

〔Example〕

A first embodiment of the present invention is shown in FIG. 1 below.

In the figure, motion vectors are searched in the following order.

1) Measure the similarity with the signal of the previous frame located at the same position as the block of the current frame. In the figure, it is indicated by a black circle. If the similarity is greater than a certain threshold, the block is treated as a block with no movement (invalid block) and the search is terminated.

2) If the degree of similarity is less than the threshold, a motion vector corresponding to a position shifted by ±4 pixels in the horizontal direction or ±4 scanning lines in the vertical direction is generated, and the degree of similarity in each is determined. The generated motion vectors are indicated by white circles in the figure. If the similarity of the black circle is the highest among the nine ways including the black circle, proceed to 3), otherwise proceed to 4).

3) Generate a motion vector corresponding to a shift of ±2 pixels in the horizontal direction or ±2 scan lines in the vertical direction.

Measure similarity. It is shown as a square in the figure. For example, if the lower right square has the highest degree of similarity, there are eight ways around it (
The similarity in the triangles in the figure is compared, and the highest one is adopted as the final motion vector for this block.

4) For example, if the white circle at the top left has the highest degree of similarity, the degrees of similarity are measured in eight ways (marked with an x in the figure) extending from there toward the top left. If the similarity of the X mark on the upper right is the highest, a motion vector is searched from that point using the same method as shown in 3).

FIG. 2 shows a second embodiment of the invention. Here, in the above search 2), the number of motion vectors for measuring similarity is increased from 8 to 12.

Instead, if the inner five ways are selected, the search in step 3) is reduced from eight ways to four, as shown by the squares in Figure 1.

As a result, the number of searches when the amount of image movement is small is the same as in the first embodiment, and since small motion vectors that occur frequently are generated in detail, the second embodiment has the feature that it can search with higher accuracy.

Next, a block configuration of a circuit for implementing the present invention will be shown. Before explaining the present invention in detail, a general motion compensation interframe coding circuit will be explained using FIG. The present invention is part of the circuit of FIG.

In FIG. 3, a TV signal input from a television camera 1 is converted into a digital signal by an A/D conversion circuit 2, and then converted into a signal in units of blocks by a block conversion circuit 3. The motion compensation circuit 4 compares the signals of the current frame and the previous frame, and selects the previous frame signal having the highest degree of similarity.

Output as a motion compensated interframe prediction signal. Furthermore, a motion vector indicating this position is sent to the encoding circuit 8.

The above motion-compensated interframe prediction signal is converted into a block signal by a block conversion circuit, and an interframe prediction error is calculated by a subtraction circuit. The prediction error signal is quantized by a quantization circuit 7. The encoding circuit 8 encodes the quantized signal and the motion vector, and sends it to the transmission 10 through the transmission control circuit 9.

At the same time, the quantized signal is restored to a prediction error signal by the inverse quantization circuit 11, and added to the above-mentioned motion compensated interframe prediction signal by the addition circuit 12 to be decoded into the original TV signal. The decoded TV signal is the same as that obtained at the receiving end. This signal is stored in the frame memory 13 and used as a "previous frame signal" when processing the next frame.

The present invention relates to the motion compensation circuit 4 shown in FIG. The detailed configuration will be explained below with reference to FIG.

The current frame signal and the previous frame signal are stored in buffer memories 21 and 22, respectively, and read out as necessary. A motion vector generation circuit 23 generates a motion vector. The shift circuit 24 is.

The signal of the previous frame is delayed in accordance with the above motion vector.

The subtraction circuit 25 calculates the difference between the current frame signal and the previous frame signal delayed by the shift circuit 24. The difference is measured by the similarity measuring circuit 26, and by the comparing circuit 27,
It is compared with the previous highest similarity of motion vectors. If the similarity is greater, the comparison circuit 27 sends out a control signal and writes the current similarity into the latch 28 and the motion vector into the latch 29.

The motion vector generation circuit 23 sends out a control signal after performing a motion vector search a predetermined number of times (for example, nine times in the first embodiment), and transfers the motion vector of the latch 29 to the latch 30 and 31. Forward. The motion vector generation circuit 23 generates subsequent motion vectors according to the output signal of the latch 31.

The adder circuit 32 uses the motion vector written in the latch 3o and the motion vector newly generated from the motion vector generation circuit 23 (for example, in the first embodiment, a motion vector search is performed from among the first nine motion vectors, i.e., a black circle). is selected, horizontal and vertical ±2) are added to adder circuit 3.
2 and sends out the shift number 24.

Thereafter, each time the density of motion vector generation is made finer, the motion vector that has been compared and determined to have the highest degree of similarity up to that point is written into the latch 30. The motion vector generation circuit 23 sends out a control signal after generating a motion vector a number of times determined by the initial motion vector written in the latch 31 and is written in the latch 29 . The motion vector with the highest similarity among the searches up to that point is transferred to the latch 30 and stopped.

The encoding circuit reads the contents of the latch 30, encodes it as a motion vector, and sends it to the transmission path. At the same time, the shift circuit 24 delays the previous frame signal in accordance with the motion vector determined in the latch 30, and sends it out as a motion compensated interframe prediction signal.

[Brief explanation of the drawing]

1 and 2 are diagrams for explaining the motion vector detection method according to the present invention, FIG. 3 is a block diagram of a general motion compensated interframe coding device, and FIG. 4 is a feature of the present invention. FIG. 2 is a block configuration diagram of a motion compensation circuit. In FIGS. 3 and 4, 1...TV camera;
2... A/D converter, 3, 5... Block conversion circuit, 4... Motion compensation circuit, 6... Subtraction circuit, 7...
Quantization circuit, 8... Encoding circuit, 9... Transmission control circuit, 10... Transmission line, 11... Inverse quantization circuit, 12
．． 32...Addition circuit, 13...Frame memory, 2
1.22...Buffer memory, 23...Motion vector generation circuit. 24... Shift circuit, 26... Similarity measurement circuit, 2
7... Comparison circuit, 28, 29, 30, 31... Latch. 1 Jaw 8 No 1 Figure L Figure 2

Claims (1)

[Claims] means for collecting a plurality of TV signals and converting them into signals in blocks; means for generating motion vectors at coarse density intervals;
means for delaying the TV signal of the previous frame for a period determined by the above motion vector; means for converting the plurality of delayed TV signals into blocks; means for measuring the degree of similarity between the above two blocks; means for comparing and determining the one with the highest degree of similarity among a plurality of motion vectors; means for generating a different number of motion vectors at high density around the motion vector determined to have the highest degree of similarity;
Means for comparing and determining the motion vector with the highest similarity among them; means for delaying the TV signal of the previous frame corresponding to the motion vector to generate a predicted signal of the TV signal to be encoded; and the TV signal to be encoded. A motion-compensated interframe coding device for a TV signal, comprising means for calculating a difference between the motion vector and the prediction signal to obtain a prediction error, and means for encoding and transmitting the motion vector and the prediction error.