JP2003299085A - Image compression device - Google Patents

Abstract

(57) [Summary] (With correction) [PROBLEMS] To provide an image compression apparatus capable of arbitrarily changing coefficients of a low-pass filter incorporated in a pre-filter and suppressing generation of useless high frequency components in compression processing. Realization. An image compression apparatus that has a pre-filter 1-2 that performs a filter process on an input image and that compresses and encodes the image data after the filter process.
The coefficient of −2 is changed in real time according to at least one of the quantization step width and the input image size to remove high frequency components.
-2, the coefficient of the low-pass filter incorporated therein can be arbitrarily changed to suppress the occurrence of useless high-frequency components in the compression processing.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to improvement of a system in which an image compression device and a prefilter are combined.

[0002]

2. Description of the Related Art In recent years, image processing technology, especially image information,
Audio information compression technology has made remarkable progress, and it has become possible to transmit high-quality moving images and audio even on transmission lines with small capacity. On the other hand, the processing capability of a processor that performs processing by software has significantly improved, and the number of cases in which the image compression processing that has been conventionally performed by hardware is realized by the processor has increased. However, due to various reasons such as price, the processing capability of the processor may not be sufficient to perform image compression. In this case, it is difficult to maintain the same image quality and frame rate as the original image. To maintain image quality, set a lower frame rate, or conversely, maintain a frame rate but set a lower image quality for compression. By performing the processing, the processing in the processor is reduced. Or, the transmission capacity of the transmission line is not enough,
When the compressed data generated by the image compression device cannot be sent, it is necessary to increase the compression rate and control the generated code amount according to the transmission path for transmission. This operation of increasing the compression rate results in setting the image quality to be maintained and the frame rate lowered, or to maintain the frame rate and setting the image quality to be reduced. Therefore, when the processing capacity of the processor is not sufficient or the transmission capacity of the transmission path is not sufficient, the image quality and the frame rate have a trade-off relationship in which only one of them is emphasized. Therefore, there is an application for image transmission in which a certain degree of image quality deterioration is allowed, such as placing importance on motion. That is equivalent to increasing the compression rate of one image in order to increase the number of frames per second.

The relationship between the image quality and the compression rate of the image compression apparatus using this processor will be described with reference to the block diagram of the image compression apparatus of FIG. First, the configuration of the image compression device will be described. The image compression device converts input image data into compressed data and outputs the compressed data. First, the input image 1-1 input to the pre-filter 2-1 is filtered and output as a converted image 1-3. The filter processing differs depending on the application, and includes, for example, a low-pass filter, image size conversion, noise removal processing, and the like. The converted image 1-3 includes a difference processing unit 1-4 and a motion detecting unit 1.
Input to -25. The converted image 1-3 and the predicted image 1-28 are input to the difference processing unit 1-4, and the converted image 1-3 is input.
On the other hand, the difference from the predicted image 1-28 is calculated, and the image data of the difference is output as the difference image 1-5. The orthogonal transform unit 1-6 performs an orthogonal transform on the input difference image 1-5 and outputs the transformed coefficient data 1-7. The orthogonal transform includes, for example, DCT and wavelet transform, transforms image data into a spatial frequency domain, and biases transform coefficients into a low frequency domain by utilizing correlation with adjacent pixels,
Data compression is performed in combination with the quantizer 1-8 in the subsequent stage. The quantizer 1-8 quantizes the input coefficient data 1-7 and outputs it as quantized coefficient data 1-9 to the variable length encoder 1-10 and the inverse quantizer 1-19. To do. The orthogonal transform unit 1-6 in the preceding stage expresses the value of the transform coefficient biased to the low frequency region by a multiple of the quantization step width, and performs data compression.

The variable length coding unit 1-10 converts the input quantized coefficient data 1-9 into a variable length code, and outputs it to the buffer 1-12 as variable length code data 1-11.
That is, a variable length code is assigned according to the probability of appearance of the symbols of the quantized coefficient data 1-9, and the average code amount of the output data is minimized. For this variable-length coding, for example, Huffman coding or arithmetic coding is used. Buffer 1-
12 temporarily stores the input variable length code data 1-11 and outputs it as compressed data 1-13. Also, the variable length code data 1 stored in the buffer 1-12
The data amount of -11 is output to the code amount control unit 1-15 as the buffer occupation amount 1-14. Code amount control unit 1-15
Is a quantization step width for controlling the compression rate of the next image from the calculation result by calculating the generated code amount when one image is compressed by the input buffer occupation amount 1-14.
It is output as -16 to the quantizer 1-8. Inverse quantizer 1
-19 is for the input quantized coefficient data 1-9,
The inverse quantization process is performed and the decoded coefficient data 1-20 is output to the orthogonal inverse transform unit 1-21. Orthogonal inverse transform unit 1-21
Performs orthogonal inverse transform processing on the input decoding coefficient data 1-20, and outputs it as a decoded image 1-22 to the frame memory 1-23. The frame memory 1-23 stores the decoded image 1-22, and as a reference image 1-24 to be used for predictive coding of the next image data, the motion detecting unit 1-23.
-25 and the motion compensation unit 1-27. Motion detector 1
-25 detects a partial image movement, pan / tilt, etc. from the converted image 1-3 and the reference image 1-24 which is the image data of the frame before the converted image 1-3,
The information is used as the motion vector 1-26, and the motion compensation unit 1
Output to -27. The motion compensation unit 1-27 performs motion compensation from the input reference image 1-24 and the motion vector 1-26, and the difference processing unit 1-4 as a predicted image 1-28 that predicts the image data of the next frame. Output to. The above is the configuration of the image compression apparatus.

Next, the operation of this image compression apparatus will be described. For the input image data 1-1, the image size is converted by the pre-filter 2-1 and the high frequency components are deleted by the low-pass filter. In the image compression, since the data compression is performed by removing the high frequency component of the image data, the low-pass filter processing before the compression is effective in improving the compression efficiency. The converted image 1-3 processed by the pre-filter 2-1 is input to the difference processing unit 1-4. The difference image 1-5, which is difference data from the prediction, is output by calculating the difference between the conversion image 1-3 and the prediction image 1-28 in which the image data of the previous frame is compressed / decompressed and motion-compensated. To do. If the prediction is correct, the difference data becomes zero, and the compression efficiency is highest. The difference image 1-5 is
The orthogonal transform unit 1-6 performs the orthogonal transform to transform the image data into the spatial frequency domain. That is, it is output to the quantization unit 1-8 as coefficient data 1-7 in which the transform coefficient is biased to the low frequency region by utilizing the correlation with the adjacent pixel. The coefficient data 1-7 is obtained by converting the value of the transform coefficient biased to the low frequency region by the orthogonal transform unit 1-6 according to the quantization step width 1-16 input to the quantization unit 1-8. 1-1
It is expressed by a multiple of 6, and the same data is made continuous by mainly widening the quantization step width of coefficient data in the high frequency region, or data is compressed by deleting it in the higher frequency region. The coefficient data biased to the low frequency region is output to the variable length coding unit 1-10 as the quantized coefficient data 1-9. The variable-length coding unit 1-10 uses the fact that the quantized coefficient data 1-9 is data biased to the low-frequency region, assigns a variable-length code according to the probability of appearance of a symbol, and outputs it. Long code data 1-1
The average code amount of 1 is minimized.

The variable-length code data 1-11 is temporarily stored in the buffer 1-12, and the variable-length code data 1-11 generated in a burst is set to a constant rate according to the transmission capacity of the transmission line. Output as compressed data 1-13. At this time, the compressed data 1-
When the data amount of 13 exceeds the transmission capacity of the transmission path, the data that cannot be transmitted is accumulated in the buffer 1-12. If the amount of generated compressed data 1-13 is small, the amount of data stored in the buffer 1-12 will decrease. By outputting the buffer occupancy 1-14 indicating the storage state of the data amount to the code amount control unit 1-15, it is possible to perform rate control for generating the data amount according to the transmission capacity of the transmission path. It is the code amount control unit 1-15 that performs rate control.
From the contents of the input buffer occupancy 1-14,
It is determined whether or not the amount of the generated compressed data 1-13 exceeds the transmission capacity of the transmission path, or vice versa. If you are generating more data than the transmission capacity,
The step width of the quantization step width 1-16 is increased to suppress the data amount of the high frequency component generated in the quantization unit 1-8. On the other hand, when the transmission capacity is not reached, the step width of the quantization step width 1-16 is reduced and the amount of data in the high frequency region generated in the quantization unit 1-8 is increased. In this way, the compression rate is controlled.

The quantized coefficient data 1-9 output from the quantizer 1-8 are the inverse quantizer 1-19 and the orthogonal inverse transformer 1.
-21, the decoded image which is the image data before compression 1-
Returned to 22. However, since the high frequency component is removed by the quantizer 1-8, it does not match the image before compression. The decoded image 1-22 is displayed in the frame memory 1-23.
And are stored for use as reference images 1-24 in motion processing. In the motion detector 1-25,
This reference image 1-24 is compared with the converted image 1-3 which is the image data of the next frame to detect a partial image motion, pan / tilt, etc.
26 to the motion compensation unit 1-27. From the motion vector 1-26 and the reference image 1-24, the motion compensation unit 1-
In 27, the predicted image 1 which is the image data of the next frame
-28 is generated and output to the difference processing unit 1-4.

As described above, in the difference processing section 1-4,
Difference data from the prediction is generated by taking the difference between the prediction image 1-28 and the conversion image 1-3. That is, if the prediction by the motion vector 1-26 is successful and the predicted image 1-28 and the converted image 1-3 match, the difference data becomes zero. In addition, the difference processing unit 1 can also be applied to a static image.
The difference is taken at -4. Here, the decoded image 1-22 that has undergone compression / decompression has the high frequency components removed, and so does the predicted image 1-28. Therefore, the converted image 1
When compared with -3, the high frequency component is output as difference data. This difference data 1-5 is stored in the quantizer 1-8.
Then, since it is output as the quantization coefficient data 1-9 of the high frequency component, the amount of generated data increases, and the buffer 1-1
The amount of data stored in 2 also increases. Since the increase in the data amount in the buffer 1-12 is output as the buffer occupation amount 1-14, the code amount control unit 1-15
Then, in order to reduce the amount of data, the quantization step width 1-
The step width of 16 is increased to try to reduce higher frequency components. In order for this series of operations to go around once,
Since the processing is performed by software, there is a delay due to the processing time, and the image in which the high frequency components are actually reduced by the quantizer 1-8 is image data at another position temporally or spatially. As a result, useless deterioration of image quality occurs. Also, if there are many high frequency components,
Since more coefficient data will be processed, the processing amount in the orthogonal transformation unit 1-6 increases. This leads to a reduction in compression efficiency.

For this reason, in order to remove unnecessary high frequency components, the low pass filter of the pre-filter 2-1 removes the high frequency components in advance. However, when the image size conversion is also performed by the pre-filter 2-1 at the same time, since the sampling frequencies of the pixels are different, various spatial frequency components are included depending on the conversion size. When the image size is changed to an arbitrary size, the optimum low-pass filter coefficients are different. When the compression rate changes, the quantizer 1
Since the range of the high frequency components deleted in -8 changes, it is not possible to completely remove useless high frequency components. Therefore, the low-pass filter that uniquely removes the high frequency components cannot completely remove the useless high frequency components, and the compression efficiency cannot be improved.

[0010]

As described above,
When the high-pass frequency component of the image data is removed by the low-pass filter built in the pre-filter, the image compression device that can arbitrarily change the image size and the compression rate cannot completely remove the useless high-pass frequency component. In addition, in the image compression device that performs compression processing by the processor, the presence of high frequency components increases the amount of processing in the quantizer and reduces compression efficiency. SUMMARY OF THE INVENTION It is an object of the present invention to eliminate these drawbacks, allow the coefficients of a low-pass filter incorporated in a pre-filter to be changed arbitrarily, and suppress the generation of wasteful high frequency components in the compression process.

[0011]

In order to achieve the above object, the present invention provides an image compression apparatus which has a pre-filter for performing a filtering process on an input image, and which compresses and codes the image data after the filtering process. The high-frequency component is removed by changing the coefficient of the pre-filter in real time. Further, the coefficient of the pre-filter is changed in real time according to the quantization step width to remove high frequency components. In addition, the coefficient of the pre-filter is changed in real time according to the input image size to remove high frequency components. In addition, the coefficient of the pre-filter is changed in real time according to the quantization step size and the input image size,
It is intended to remove high frequency components. In other words, the coefficient of the low-pass filter built in the pre-filter can be changed arbitrarily, and the coefficient of the low-pass filter is obtained using the information from the code amount control unit, so that the high-frequency component that is wasted in the compression process is generated Can be suppressed.

[0012]

1 is a block diagram showing the configuration of an embodiment of an image compression apparatus with an adaptive prefilter according to the present invention. Compared with the conventional configuration of FIG. 2, a filter coefficient control unit 1-17 is added, and a pre-filter 2-
1 has been changed to the adaptive pre-filter 1-2,
The configuration is the same as the conventional one. The rate control operation for determining the data amount and image quality of the generated compressed data is similar to the conventional one, and the data amount in the buffer 1-12 is output to the code amount control unit 1-15 as the buffer occupation amount 1-14, and the The quantization step width 1-16 is output to the quantizer 1-8. The present invention is different from the conventional one in that the quantization step width 1-16 is also output to the filter coefficient control unit 1-17.

The filter coefficient control unit 1-17 determines the range in which the high frequency components are removed by the input quantization step width 1-16. From the contents of this determination, the coefficient of the low-pass filter for removing the high frequency components is determined, and the adaptive pre-filter 1 is used as the filter coefficient 1-18.
Output to -2. At the same time, the size of the image that is being compressed is input as image size information 1-29, and the filter coefficient that matches the sampling frequency of the image data is determined. The adaptive pre-filter 1-2 can input the filter coefficient 1-18 for changing the coefficient of the built-in low-pass filter. That is, in the filter coefficient control unit 1-17, the quantization step width 1-
16 and the filter coefficient 1-18 derived from the image size information 1-29, by changing the coefficient of the low-pass filter of the adaptive pre-filter 1-2 in real time,
It becomes possible to remove high frequency components that match the current compression rate and image size, and the wasteful high frequency components are quantized by the quantizer 1-
It is possible to perform compression processing without inputting in 8.
As a result, the amount of processing in the quantization unit 1-8 by the processor can be reduced and the compression efficiency can be improved.

[0014]

By using the configuration of the present invention described above, it is possible to remove high frequency components that match the current image size and compression rate, and even if the image size and compression rate are changed arbitrarily, there is no waste. It is possible to realize an image compression device that can remove various high frequency components.

[Brief description of drawings]

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

FIG. 2 is a block diagram showing an example of the configuration of a conventional image compression device.

[Explanation of symbols]

1-1: Input image, 1-2: Adaptive pre-filter, 1-
3: Transformed image, 1-4: Difference processing unit, 1-5: Difference image, 1-6: Orthogonal transformation unit, 1-7: Coefficient data, 1-
8: quantizer, 1-9: quantized coefficient data, 1-10:
Variable length coding unit, 1-11: Variable length code data, 1-1
2: buffer, 1-13: compressed data, 1-14: buffer occupation amount, 1-15: code amount control unit, 1-16: quantization step width, 1-17: filter coefficient control unit, 1-1
8: filter coefficient, 1-19: dequantizer, 1-20:
Decoding coefficient data, 1-21: Orthogonal inverse transform unit, 1-22:
Decoded image, 1-23: Frame memory, 1-24: Reference image, 1-25: Motion detection unit, 1-26: Motion vector, 1-27: Motion compensation unit, 1-28: Predicted image, 1-
29: Image size information.

Claims (4)

[Claims] 1. An image compression apparatus having a pre-filter for performing a filtering process on an input image, wherein the pre-filter coefficient is changed in real time in an image compression apparatus for compression-encoding the filtered image data. An image compression device characterized by removing components. 2. The image compression apparatus according to claim 1,
An image compression apparatus, wherein the coefficient of the pre-filter is changed in real time according to the quantization step width to remove high frequency components. 3. The image compression apparatus according to claim 1,
An image compression apparatus, wherein the coefficient of the pre-filter is changed in real time according to an input image size to remove high frequency components. 4. The image compression apparatus according to claim 1, wherein
An image compression apparatus characterized in that a coefficient of the pre-filter is changed in real time according to a quantization step width and an input image size to remove a high frequency component.