JPH06233267A - Image data compression device - Google Patents

Abstract

PURPOSE:To provide the picture data compressing device of high encoding efficiency and a small circuit scale. CONSTITUTION:This picture data compressing device is constituted as follows. A reference frame is difference-processed by a subtracter 102 through a band dividing.thinning processing circuit 101, and is given re-quantization corresponding to a visual characteristic by a requantizer 103, and is variable-length-encoded by a variable-length encoder 104, and is inverse-quantized by an inverse quantizer 105, and is added to a previous reference frame by an adder 106, and is stored in a frame memory 107, and its data is compared with the data encoded next successively from a low frequency band side toward a high frequency band side by a motion compensation circuit 108.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image data compression apparatus used for communication systems and storage media systems (disks, tapes, etc.) that require band compression of images.

[0002]

2. Description of the Related Art Recently, in order to realize a high S / N, many attempts have been made to digitize image data and record or transmit it. However, digitization of image data for recording and reproduction requires a wide band, resulting in an increase in required area of the medium, an increase in cost, and an increase in communication cost. Therefore, attempts have been made to significantly reduce the amount of image data by utilizing the correlation that images have and the visual characteristics that humans have.

A conventional image data compression device will be described below. FIG. 4 is a block diagram showing the configuration of a conventional moving image data compression apparatus by band division. Since the image data has a large correlation between frames, the subtractor 401
Thus, the subtraction process between the frames is performed using the frame image before or after the frame image as the reference frame. The band-divided data is subjected to band division and thinning processing in the band division / thinning processing circuit 402. In the case of an image, the band division and the thinning process are two-dimensionally performed in both the horizontal direction and the vertical direction. As shown in FIG. 5, band division is performed in the horizontal direction using a low-pass filter circuit 501 and a high-pass filter circuit 502. Then, each of the band-divided data is sub-sampled by the sub-sampling circuit 503 in a 2: 1 ratio.

Similarly, the data sub-sampled in the horizontal direction is band-divided by the low-pass filter and the high-pass filter in the vertical direction and sub-sampled. Here, for the data on the low frequency side in the horizontal direction and the vertical direction, band division and sub-sampling are recursively repeated. FIG. 6 shows how the band division is performed when it is repeated three times. In FIG. 6, fs represents the sampling frequency of the original image. By the first band division, it is divided into two with fs as the center, the low frequency side is further divided into two with fs / 8 as the center, and the low frequency side is fs.
It is divided into two around / 16. In this way, each band is divided into octaves in order to recursively divide the band into halves on the low frequency side. In FIG. 6, three times,
Although band division and subsampling are recursively repeated, the number of repetitions may be arbitrary. FIG. 7 shows a two-dimensional representation of the data obtained by band division and subsampling. Here, the left side shows an image of a low-frequency side area in the horizontal direction, and the right side shows an image of a high-frequency side area in the horizontal direction. Further, the upper side shows an image of a region on the low frequency side in the vertical direction, and the lower side shows an image of a region on the high frequency side in the vertical direction.
Next, the recursively divided data is requantized by the requantizer 403 according to each region (band). Here, the requantization performed is quantized according to the sensitivity of human contrast to each band. In general, the low frequency side is finely quantized, and the high frequency side is roughly quantized. The quantized data of each area is variable-length coded by the variable-length encoder 404. A Huffman code, an arithmetic code, or the like is used as a variable length coding method. Since re-quantized data are concentrated near zero, a short code length is generally assigned to values near zero and a long code length is assigned to values with large absolute values. assign. On the other hand, since the encoded data is used as a reference frame of an image to be encoded later, it is reconstructed. Therefore, the data requantized by the requantizer 403 is dequantized by the dequantizer 405.
The inversely quantized data is interpolated by the interpolation processing circuit 406. In the interpolation processing, contrary to the processing performed by the band division / thinning processing circuit 402, the low frequency band and the high frequency band are paired in order from the low frequency side, and are added. The interpolated data is added to the previous reference frame used when encoding. The reconstructed image data is accumulated in the frame memory 408 to be used as a reference frame of an image to be encoded next. Then, the accumulated reference frame is subjected to motion vector calculation and motion compensation by the motion compensation circuit 409 in order to enhance the correlation with the frame to be encoded. FIG. 8 is a schematic diagram for explaining a conventional motion vector calculation method. The calculation method calculates the addition of the absolute value of the difference between the block of the calculation area having an arbitrary number of pixels of the image to be encoded and the block of the calculation area having an arbitrary number of pixels of the reference image over a certain search range. Then, the shift of the pixel having the minimum value becomes the motion vector. This is generally called a block matching method. The size of the calculation area for performing the matching calculation and the search range for obtaining the motion vector may be arbitrary. In this way, the reference frame is motion-compensated according to the obtained motion vector.

[0005]

However, in the above-described conventional configuration, since the matching is performed in block units, if there is a motion other than parallel movement between frames, such as rotation or deformation, the correlation between frames is significantly reduced. .

In addition, reverse conversion processing (interpolation processing) is required to reconstruct the reference frame.

The present invention solves the above-mentioned conventional problems. In the moving picture coding system, the correlation between frames is improved, the coding efficiency is improved, and the inverse converter (interpolation processing) is used. It is an object of the present invention to provide an image data compression device that does not require and reduces the circuit scale.

[0008]

In order to achieve this object, the image data compression apparatus of the present invention recursively performs band division, then hierarchically obtains motion vectors from the low frequency side, and The difference processing between the two is configured to be performed on the image subjected to the band division / thinning processing.

[0009]

With the above-described structure, the present invention operates so that the correlation between frames is enhanced, the coding efficiency is improved, and the inverse conversion process is unnecessary.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing the arrangement of an image data compression apparatus according to this embodiment. In FIG. 1, the image data is first subjected to band division and thinning processing in a band division / thinning processing circuit 101. In the case of an image, the band division and the thinning process are two-dimensionally performed in both the horizontal direction and the vertical direction. As shown in FIG. 5, band division is performed in the horizontal direction using a low-pass filter circuit 501 and a high-pass filter circuit 502. And
Each of the band-divided data is sub-sampled by the sub-sampling circuit 503 in a 2: 1 ratio.
Here, as the low-pass filter and the high-pass filter, filters that are completely restored in the decoding process (interpolation process and filtering process) are selected.

The data sub-sampled in the horizontal direction is also band-divided by the low-pass filter and the high-pass filter in the vertical direction and sub-sampled. Here, for the data on the low frequency side in the horizontal direction and the vertical direction, band division and sub-sampling are recursively repeated. FIG. 6 shows how the band division is performed when it is repeated three times. In FIG. 6, fs represents the sampling frequency of the original image. By the first band division, it is divided into two with fs as the center, the low frequency side is further divided into two with fs / 8 as the center, and the low frequency side is fs.
It is divided into two around / 16. In this way, each band is divided into octaves in order to recursively divide the band into halves on the low frequency side. In FIG. 6, three times,
Although band division and subsampling are recursively repeated, the number of repetitions may be arbitrary. FIG. 7 shows a two-dimensional representation of the data obtained by band division and subsampling. Here, the left side shows an image of a low-frequency side area in the horizontal direction, and the right side shows an image of a high-frequency side area in the horizontal direction. Further, the upper side shows an image of a region on the low frequency side in the vertical direction, and the lower side shows an image of a region on the high frequency side in the vertical direction.

The moving image data is subjected to a difference processing between frames by using a band-divided image before or after the subtracter 102 as a reference frame. The requantizer 103 requantizes the difference-processed data according to each region (band). Here, the requantization performed is quantized according to the sensitivity of human contrast to each band. In general, the low frequency side is finely quantized, and the high frequency side is roughly quantized. The quantized data of each area is variable-length coded by the variable-length encoder 104. A Huffman code, an arithmetic code, or the like is used as a variable length coding method. The requantized data generally have a frequency of occurrence near zero, so
Short code lengths are assigned to values near zero, and long code lengths are assigned to values with large absolute values. On the other hand, in order to use the encoded data as a reference frame of an image to be encoded later, it is reconstructed by performing a process reverse to the encoding. Therefore, the data requantized by the requantizer 103 is dequantized by the dequantizer 105. The dequantized data is added to the adder 106.
Is added to the previous reference frame used when encoding. The reconstructed image data is used as a reference frame of an image to be encoded next, so that the frame memory 1
It is stored in 07.

Further, the accumulated reference frame is subjected to motion vector calculation and motion compensation by the motion compensation circuit 108 in order to improve the correlation with the frame to be encoded. FIG. 2 is a schematic diagram for explaining a relationship between band-divided data to be encoded and band-divided data that is referred to when a motion vector is obtained. The numbers in the figure are numbers corresponding to the respective regions shown in FIG. 7, and show that the regions are from the low frequency region to the high frequency region as going from top to bottom. Now, in order to hierarchically obtain the motion vector from the low frequency side to the high frequency side, first of all, in the encoded band division data,

[0015]

[Outer 1]

Band division data referred to as

[0017]

[Outside 2]

Are compared, and a motion vector is obtained for each pixel of the area (outer 1). Here, the search range of the motion vector may be arbitrary. In addition, the area for comparison calculation is
It is one pixel. In the same way, toward the high frequency side,
The motion vector is obtained hierarchically. FIG. 3 is a schematic diagram illustrating a method of hierarchically obtaining motion vectors.

[0019]

[Outside 3]

Since the area (outer 1) is doubled in the data of the area (2), the point obtained by doubling the motion vector obtained in (outer 1) is used as a reference (outer 3). The motion vector of the region is calculated. The search range of the motion vector in the area (outer 3) is calculated in the range of 0 to 1 pixel in both the horizontal and vertical directions, and the absolute value is calculated in pixel units to obtain the motion vector. region

[0021]

[Outside 4]

The motion vector of is also obtained in the same manner.
The obtained motion vector value has a high correlation in each area, and thus the difference processing is performed in each area. Since the data of the difference-processed motion vector values are concentrated near zero, variable-length coding is performed in which a short code is assigned to values near zero, and the data is sent. Next, motion compensation of the reference frame is performed according to the obtained motion vector. region

[0023]

[Outside 5]

With respect to, the motion compensation of the reference frame is performed according to the motion vector obtained in the area (outer 1). Also the area

[0025]

[Outside 6]

In the step (3), the motion compensation of the reference frame is performed according to the motion vector obtained in the area (outer 3). Similarly,
region

[0027]

[Outside 7]

In, the motion compensation of the reference frame is performed according to the motion vector obtained in the area (outer 4). And
Motion-compensated band-divided data as a reference frame,
The data of the band-split frame to be encoded is the subtractor 102.
Is subjected to difference processing and encoded.

As described above, according to the present embodiment, the image data is band-divided and sub-sampled, the difference from the reference frame is obtained by the subtractor, the re-quantizer is re-quantized according to the human visual characteristics, and the variable length is changed. Encode. On the other hand, inversely quantized and added to the previous reference frame, sequentially compared with the data to be encoded next using the motion compensation circuit from the low frequency side to the high frequency side, hierarchical motion compensation, Since the band division frame data is encoded, the encoding efficiency is improved.

[0030]

As is apparent from the above-described embodiments, according to the present invention, in a compression device for encoding an image by band division, after the band division, a motion vector is hierarchically obtained from the low frequency side to obtain a frame. Provided is an image data compression device capable of increasing the correlation between channels and improving the coding efficiency. Further, since the data after band division is compared, the inverse conversion process of the reference frame is not necessary, and the circuit scale can be reduced. it can.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of an image data compression apparatus according to an embodiment of the present invention.

FIG. 2 is a schematic diagram for explaining a relationship between band-divided data to be encoded and band-divided data referred to when a motion vector is obtained in the same embodiment.

FIG. 3 is a schematic diagram for explaining a method for hierarchically obtaining motion vectors in the same embodiment.

FIG. 4 is a block diagram showing a configuration of a conventional image data compression device.

FIG. 5 is a block diagram showing a configuration of a band division and thinning processing circuit in a conventional example.

FIG. 6 is a characteristic diagram for explaining band division in a conventional example.

FIG. 7 is a schematic diagram for explaining a band division region in a conventional example.

FIG. 8 is a schematic diagram for explaining a method for obtaining a motion vector in a conventional example.

[Explanation of symbols]

 101 Band Division / Decimation Processing Circuit 102 Subtractor 103 Requantizer 104 Variable Length Encoder 105 Inverse Quantizer 106 Adder 107 Frame Memory 108 Motion Compensation Circuit

Claims (3)

[Claims] 1. A band division / decimation processing circuit for dividing image data into a low band and a high band and thinning out band-divided data, and data processed in the band division / thinning process circuit from a reference frame. A subtractor that performs difference processing, a requantizer that requantizes the difference processed data, a variable length encoder that performs variable length coding on the requantized data, and the requantized To make the data the next reference frame, an inverse quantizer for dequantizing, an adder for adding the dequantized data, and the previous reference frame used for encoding, A frame memory for storing the added data as a reference frame until the next encoding, and a motion vector calculation and motion compensation between the reference frame and the frame to be encoded. Image data compression apparatus and a motion compensation circuit for performing. 2. The image data compression apparatus according to claim 1, wherein the motion compensation circuit operates so as to obtain a motion vector hierarchically in order from the low band side for each band-divided region. 3. The image data compression apparatus according to claim 1, wherein the value of the motion vector in the same layer is subjected to difference processing with respect to the value of the motion vector obtained previously.