JP3383450B2 - Digital image ringing noise eliminator - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a ringing noise removing apparatus capable of removing ringing noise in a high frequency region of a digital image decompressed from image compressed data.

[0002]

2. Description of the Related Art Over the past few years, MPEG-1,
MPEG-2, JPEG, H.264. Many image compression methods have been proposed and standardized for the transmission of digital images such as H.261 and other proprietary standards. Most compression standards achieve high compression rates, thus effectively losing a lot of data. This much loss causes problems when the compressed digital image data is decompressed to reconstruct the original digital image. To reduce or eliminate these inconveniences,
Post processing of the decompressed digital image is required.

[0003]

However, most of the current post-processing methods are IIR (Infinite Impulse Response).
Alternatively, a filter based on FIR (finite impulse response) is used. The implementation of this filter is complex and costly, and it also processes image data that is noise-insensitive, which causes a loss of reconstructed image quality. In particular, this filter has a problem that ringing noise is generated around the high frequency region of the reconstructed digital image.

Therefore, in consideration of such conventional problems, the present invention removes ringing noise around the high frequency region of an image reconstructed from digital compressed image data, and retains high frequency details. SUMMARY OF THE INVENTION It is an object of the present invention to provide a ringing noise elimination device for a digital image, which can prevent deterioration of image quality.

[0005]

According to a first aspect of the present invention, there is provided image decoding means for decoding compressed digital image data and outputting reconstructed image pixel data composed of blocks of a plurality of pixels. Ringing noise detecting means for detecting the presence or absence of ringing noise for each block on the basis of a predetermined variable threshold value in the reconstructed image pixel data, state storing means for storing the state of the detection, and A ringing noise removing device for a digital image, comprising: ringing noise removing means for smoothing the ringing noise from the reconstructed image pixel data by using a state for smoothing control.

According to a second aspect of the present invention, the ringing noise removing means detects ringing noise for each block in the raster-scanned reconstructed image pixel data to obtain a predetermined variable threshold value. On the basis of the selected pixel to be smoothed, the smoothing filter output of the selected pixel is calculated, and the calculated pixel or the reconstruction is performed based on the detection state of ringing noise for each block. 2. The digital image ringing noise elimination device according to claim 1, wherein either one of the image pixel data thus generated is selectively output.

[0007]

With the above arrangement, the apparatus of the present invention reads the reconstructed image pixel data block by block from the decompressor (or decoder). Then each block
High frequency regions are detected that are potential sources of ringing noise. The frequency level detected depends on the detection threshold that can be adjusted. Next, the detected block detection state is stored. This detection state is used for smoothing control. Meanwhile, the reconstructed image pixel data is converted from block format to raster scan by the reorder memory. The pixel data is converted into a raster scan and then output to a ringing noise filter. Pixels to be filtered are 3x
It is assumed to be at the center of the 3 matrix. Next, every pixel in this matrix is compared to an adjustable threshold. Pixels within a range are then selected for the calculation of the smoothing result. The selected pixel is then used to calculate the smoothing result for the particular pixel. The smoothed pixel is then replaced with the original pixel using the pre-stored detection state. This detect state is used to control the multiplexer that selects the final output.

[0008]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 7A is a diagram showing an example of the frequency response in the high frequency region of the original digital image, and FIG. 7B is a ringing noise near the high frequency region reconstructed from the compressed image data. It is the figure which showed. The present invention is for removing this ringing noise.

FIG. 1 is a block diagram of an embodiment of a digital image ringing noise removing apparatus according to the present invention. That is, the image decoder 1 is a device for restoring compressed image data. Ringing noise detector 2
Is a device for detecting ringing in the reconstructed image. The state memory 3 is a memory for storing the detection state of the ringing noise detector 2. The reorder memory 4 is a memory for converting pixel data in a block format into raster scan and for other decoder systems. The ringing noise filter 5 has
Pixels to be filtered and pixels not to be filtered are classified, and only the pixels to be filtered are processed to remove ringing noise. The interpolation / post-processing device 6 is a device that performs other post-processing before the final post-processing is performed and the reconstructed digital image is output.

Next, the operation of this embodiment will be described.

When the compressed image data 40 is input to the image decoder 1 to reconstruct the original image, it is decompressed and output as reconstructed image pixel data 41. The reconstructed image pixel data 41 output from the image decoder 1 is in a block format (for example, 8 × 8 pixels). This output is reorder memory 4
And ringing noise detector 2. The reorder memory 4 rearranges the frames of the image decoder 1 and converts the block format data into a raster scan format.

The ringing noise detector 2 will be described in more detail with reference to FIG. Since the output from the image decoder 1 is random and has a high transfer rate, the reconstructed image pixel data 41 output from the image decoder 1 shown in FIG. 1 is stored in the FIFO buffer 7 (first-in first-out memory). Is entered. The detection controller 8 is a FI indicating the state of the FIFO buffer 7.
The free state flag of the FO buffer 7 is checked. FI
When the FO buffer 7 has data, the detection controller 8 reads the data from the FIFO buffer and supplies it to the maximum pixel comparator 9, average block calculator 10 and minimum pixel comparator 11. The output of this FIFO buffer 7 is read into the maximum pixel comparator 9, average block calculator 10 and minimum pixel comparator 11 for calculation.

The maximum pixel comparator 9 checks and compares all the pixels included in the reconstructed block in order to detect the maximum value while using the read signal as an output clock. The maximum pixel comparator 9 outputs MAX 42 which is one of the inputs of the threshold comparator 12. The minimum pixel comparator 11 checks and compares all the pixels included in the reconstructed block in order to detect the minimum value while using the read signal as an output clock. The minimum pixel comparator 11 outputs the MIN 44 which is one of the inputs of the threshold value comparator 12. The average block calculator 10 calculates the average value of the pixels of the block. The average block calculator 10 outputs the MEAN 43 which is one of the inputs of the threshold value comparator 12. When the detection controller 8 reads one block of pixel data from the FIFO buffer 7, the read cycle is stopped and the output data of the maximum pixel comparator 9, average block calculator 10 and minimum pixel comparator 11 at that time is stored in the block. It is the final output. The threshold comparator 12 performs a final calculation that determines the ringing noise detection state of the block being processed based on the inputs of the MIN 44, MAX 42 and MEAN 43.

The threshold comparator 12 will be described in more detail with reference to FIG. In the subtraction circuit 13, M
The value of EAN43 is always less than or equal to MAX42, so MA
MEAN43 is subtracted from X42. In the subtraction circuit 14, the value of MIN44 is always less than or equal to MEAN43, so MIN44 is subtracted from MEAN43. The outputs from the subtraction circuit 13 and the subtraction circuit 14 are compared by the comparator 15. Then, the larger one is selected as the DC output,
It is output to the comparator 17 as DC45. Comparator 17
Then, the DC 45 is compared with the threshold value of the detection threshold value generation circuit 16. If the DC 45 is greater than the threshold of the detection threshold generation circuit 16, it is determined that the block of pixels being processed has a high frequency region and is to be filtered. And DS1 showing the detection state
(46) is output from the comparator 17. This output DS
1 (46) is the output of the threshold value comparator 12 shown in FIG. 2 and the ringing noise detector 2 shown in FIG. DS1 (4
6) is stored in the state memory 3 shown in FIG. 1 together with the write signal WR47 output from the detection controller 8 as shown in FIG.

Referring back to FIG. 1, when the pixel data included in the block format of the reorder memory 4 is converted into the raster scanning format, the pixel data is sent to the ringing noise filter 5.

Here, the ringing noise filter 5 will be described in detail with reference to FIG. The ringing noise filter 5 is synchronous and clocked by the image clock. The line memories 18 and 19 have a memory length equal to the number of pixel data of one raster scanning line. D
The FFs 24, 21, 25 and 26 are 8-bit registers. The pixel data to be filtered is the central pixel of the filter matrix. That is CP
The output of the DFF 24 is designated by X48. Pixels contributing to the calculation of the filtering of the center pixel were delayed one line before the center pixel at the output of DFF 26, one line before the clock at the output of DFF 25, and one clock at the output of line memory 18. One line and one delayed line at the output of DFF 21. All these pixels pass through the PMUXs 22, 23, 27 and 28, and the filtering calculations will be performed. These PMUXs 22, 23, 27 and 28 are
It has the same circuit configuration.

Here, PMUX 22, 23, 27 and 2
8 will be described in detail with reference to FIG. When the input pixel 49 is input to the difference absolute value calculation circuit 34, the absolute value of the difference with respect to the central pixel CPX48 is calculated. The output of the difference absolute value calculation circuit 34 is compared by the comparator 36 with the threshold value preset in the filter threshold value generation circuit 33. When the output of the difference absolute value calculation circuit 34 is larger than the threshold value of the filter threshold value generation circuit 33, the output of the multiplexer 35 becomes the input pixel 49, and at the same time, the output of the comparator 36 becomes “1”. In other cases, the output of the multiplexer 35 is set to a zero value and the output of the comparator 36 becomes "0" at the same time. In addition, as shown in FIG.
The outputs of the comparators 36 of X22, 23, 27 and 28 are connected to the divisor generator 31. Therefore, PMUX22, 2
3, 27 and 28 each have two outputs, and the output of the multiplexer 35 is sent to the sum calculation circuit 29 and the comparator 3
The output of 6 is output to the divisor generator 31. The outputs of the divisor generator 31 and the sum calculation circuit 29 are input to the special division circuit 32.

The logic and configuration of the divisor generator 31 and the special division circuit 32 are shown in FIGS. 6 (a) and 6 (b), respectively.
The divisor generator 31 generates "1" or "2" according to the number of "1" s in the four input signals according to the logic table of FIG.
Alternatively, “4” is output as the divisor 52. Special division circuit 3
2, as shown in FIG. 6B, the sum output 51 which is the output of the sum calculation circuit 29 (shown in FIG. 4) and the central pixel CPX output from the DFF 24 (shown in FIG. 4).
The value of 48 is added by the adder 37. A bit selected from either the output of the adder 37 or the sum output 51 according to the logic table of FIG. 6B while being controlled by the divisor 52 output from the divisor generator 31 (shown in FIG. 4). Is output as shown in FIG. 6B to perform the division.

As shown in FIG. 4, this special division circuit 32
Is output from the data multiplexer 30 by the DS2 (50) indicating the detection state, which is the output of the DFF 24 CP.
It is multiplexed with X48. The DS2 (50) signal is read from the state memory 3 (shown in FIG. 1) by the circular buffer control 20. That is, the circular buffer control 20 generates a memory address of the state memory 3 and a read signal to the state memory 3 which has a function of tracking the position of the pixel being filtered. Read out DS2 (50). If the state of DS2 (50) indicates the presence of ringing noise within the pixel boundaries of CPX 48, data multiplexer 30 outputs the data from special divide circuit 32. Otherwise, the pixel value will be the unchanged value before it was input to the ringing noise filter 5. The output of the data multiplexer 30 is the final output of the ringing noise filter 5.

[0021]

As described above, according to the present invention, the sharpness of the reconstructed image can be maintained, while the source causing such noise is detected and applied to the detected area. The effect is that ringing noise can be removed from the reconstructed image by applying a filter. Also, without the need for complicated mathematical calculations,
Together with the image decoder (ie decompressor) it has the advantage that it can be easily implemented independently.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of a digital image ringing noise removing apparatus according to the present invention.

2 is a configuration diagram of a ringing noise detector 2 of FIG.

FIG. 3 is a block diagram of the threshold value comparator 12 of FIG.

FIG. 4 is a block diagram of a ringing noise filter 5 of FIG.

5 is a block diagram of the PMUX 22, 23, 27 and 28 of FIG.

6 is a diagram showing the logic and configuration of a special division circuit 32 of FIG.

7A is a diagram showing an example of a frequency response of a high frequency region of an original digital image, and FIG. 7B is a diagram showing a ringing noise near the high frequency region reconstructed from compressed image data.

[Explanation of symbols]

1 image decoder 2 Ringing noise detector 3 state memory 4 reorder memory 5 Ringing noise filter

─────────────────────────────────────────────────── ─── Continuation of front page (56) References JP-A-3-276916 (JP, A) JP-A-2-56179 (JP, A) JP-A-5-183778 (JP, A) (58) Field (Int.Cl. ⁷ , DB name) H04N 5/21 H04N 7/24

Claims (2)

(57) [Claims] 1. An image decoding means for decoding compressed digital image data to output reconstructed image pixel data composed of blocks of a plurality of pixels; A ringing noise detection unit that detects the presence or absence of ringing noise for each block based on a variable threshold, a state storage unit that stores the detection state, and a detection state that is used for smoothing control. A ringing noise removing device for a digital image, comprising: ringing noise removing means for smoothing the ringing noise from the constructed image pixel data. 2. The ringing noise removing means detects ringing noise for each block in the raster-scanned reconstructed image pixel data, and smoothes the ringing noise based on a predetermined variable threshold. A specific pixel to be selected, a smoothing filter output of the selected pixel is calculated, and the calculated pixel or the reconstructed image pixel data is calculated based on the detection state of ringing noise for each block. 2. The digital image ringing noise elimination device according to claim 1, wherein either one is selectively output.