JP2761077B2 - Post filter - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a configuration of a signal processing type post filter for removing a noise component from an image signal transmitted with a high compression code.

(Prior Art) FIGS. 4 and 5 show examples of such a circuit called a conventional post filter. In FIG. 4, 1 is an image signal input terminal, 2 is an encoding device, 3 is a transmission path, 4 is a decoding device, 5 is a decoded image signal terminal, 6 is a filter block, 7
Is an image signal output terminal.

Reference numerals 61, 62, and 63 of the filter block 6 denote registers for delaying line-sequentially scanned image signals by one pixel, 64 are weighted adders, 65 and 66 are line memories that delay image signals by one scanning line, and 67 Is a weighted addition circuit.

The image signal input from the image signal input terminal 1 is sent out to the transmission line 3 which has been highly compressed and encoded by the encoding device 2. The signal received from this transmission path is decoded by the decoding device 4 and output to the decoded image signal terminal 5 as a decoded image signal.

Although the output of the decoded image signal terminal 5 can be used as a received image, in the case of high-compression coded transmission, a process called post-filtering is performed by the filter block 6 because distortion and noise generated in the process of coding and decoding are large. Do. This filter block, that is, the post-filter output is output from the image signal output terminal 7.

Next, before explaining the operation of the post-filter by the filter block 6, what kind of distortion and noise occurs under the assumption of the encoding process will be described. For example, CCITT
A scheme such as motion compensation predictive DCT transform coding indicated by the standardization recommendation H.261 or the like is applied. In this encoding method, an image signal of one frame is divided into blocks of 16 × 16 pixels or 8 × 8 pixels, and a difference from an image signal transmitted one frame before is encoded.

This encoding is performed so that only the changed blocks of the image are updated, and the unchanged or small-changed blocks retain the value of the previous frame. The change value is DCT-transformed for each block, and the DCT-transformed transform coefficient is quantized and transmitted. Here, the DCT transform is a kind of orthogonal transform for transforming image signal data into its spatial frequency component data, and the quantization approximates transform coefficients having continuous values to a finite number of states.

As a result of the encoding, various distortions and noises are generated. For example, as a result of quantization, a part of the spatial frequency component is transmitted, and one part is approximated to 0 and deleted, thereby generating ringing-like noise near a steep edge of the image. In addition, non-continuous blocks that are not present in the original image at adjacent block boundaries can be obtained by contacting blocks that are not updated by encoding and blocks that are not updated, and by random occurrence of error noise generated by quantization between blocks. Steps occur.

The purpose of the post filter is to remove these artifacts. In the filter block 6 in FIG. 4, the above-described minute ringing around the image edge and the step between blocks are removed by a low-pass filter effect.

That is, the signal input from the decoded image signal terminal 5 is sequentially input to the registers 61, 62, and 63. The values of three pixels in the horizontal direction that are line-sequentially scanned are developed in the registers 61, 62, and 63, and the values are multiplied by coefficients 1/4, 1/2, and 1/4, respectively, to obtain a weighted addition circuit 64.
Add with As a result, a signal subjected to a low-pass filter in the horizontal direction is output from the weighted addition circuit 64.

Next, the values of three pixels in the vertical direction are developed using the line memories 65 and 66. This is multiplied by coefficients 1/4, 1/2, and 1/4, respectively, and added by the weighted addition circuit 67. As a result, the signals subjected to the low-pass filter in the horizontal and vertical directions are output from the weighted addition circuit 67 to the image signal output terminal 7. This two-dimensional low-pass filter smoothes minute ringing around edges and steps between blocks to reduce distortion.

FIG. 5 shows another example of the circuit of the conventional post filter, and FIG. 4 shows a delay circuit which does not filter the filter output by the parameter signal input from the decoding device 4 via the parameter signal input terminal 8. 81 outputs and selection circuit
The difference is that it can be selected depending on 80.

In this circuit example, in order to prevent the image from being blurred due to the effect of the two-dimensional filter on the part without minute ringing or block level difference, use a parameter signal indicating the degree of change of the image and apply the filtered signal. And a signal given only a delay equal to that of the filter.

(Problems to be Solved by the Invention) The post-filters shown in the filter block 6 of FIG. 4 or FIG. 5 described above operate as a two-dimensional low-pass filter or an adaptive two-dimensional low-pass filter, respectively. The order is limited to 3 × 3.

For this reason, it is not possible to remove the distortion in the spatially expanded range. In addition, the above-described step difference between blocks is only smoothed in a range of three pixels, and cannot be sufficiently removed. One means for solving this problem is to increase the order of the filter from 3.times.3 to 5.times.5 or the like. This cannot be said to be an effective means since noise cannot be removed without causing the noise.

(Object of the Invention) It is an object of the present invention to provide a post filter capable of adaptively increasing the order of a filter in a small scale in order to remove a noise component from a highly compressed and transmitted image signal. It is in.

(Means for Solving the Problems) In order to achieve the above objects and objects, the present invention provides a signal processing method for removing a noise component from an image signal transmitted with a high compression code, in particular, a terminal for inputting a decoded image signal, A parameter signal input terminal for inputting a parameter signal indicating whether the value at each position of the decoded image has changed with respect to the value of the decoded image of the previous frame together with the decoded image signal, and a filter block output described later as a final output to the outside In the configuration of a post filter having an image signal output terminal, a filter block that conditionally selects and inputs an input decoded image signal or an after-mentioned frame memory output image signal, and a frame memory that conditionally inputs and writes the output of the filter block, The input condition of the filter block and the input condition of the frame memory according to the history of the parameter signal And a control circuit that determines each of them.

(Operation) The present invention enables the decoded image signal to pass through the two-dimensional low-pass filter a plurality of times by using the frame memory, and to adapt the number of times of passing the filter to temporal and spatial changes of the image. To remove noise components from the image signal.

(Embodiment) FIG. 1 is a block diagram showing a circuit configuration of an embodiment of the present invention. The same parts as those shown in FIGS. 4 and 5 are denoted by the same reference numerals and description thereof will be omitted.

In the figure, reference numeral 68 denotes an on / off selection circuit for a horizontal one-dimensional filter, reference numeral 69 denotes an on / off selection circuit for a vertical one-dimensional filter, and reference numeral 82 denotes control for determining input conditions of a filter and input conditions of a frame memory based on a history of parameter signals. circuit,
83 is a frame memory, 84 is a filter input selection circuit, 85 is a frame memory input write enable signal, and 86 is a filter input selection signal.

The decoded image signal is obtained at the decoded image input terminal 5 as in the conventional circuit example. This decoded image signal is connected to a first input of a filter input selection circuit 84. The filter input selection circuit 84 is controlled by a filter input selection signal 86 output from the control circuit 82. That is, when the decoded image signal changes from the previous frame, the first input decoded image signal is selected, and when the decoded image signal does not change from the previous frame, the output of the second frame memory 83 is selected.

The output of the filter input selection circuit 84 is input to the next horizontal one-dimensional low-pass filter. This horizontal one-dimensional low-pass filter operates similarly to the conventional circuit example. The output of the horizontal one-dimensional low-pass filter is input to the horizontal one-dimensional filter on / off selection circuit 68. The output of the next-stage vertical one-dimensional low-pass filter is input to an on / off selection circuit 69 for the vertical one-dimensional filter.

The two selection circuits 68 and 69 perform a filtering process on the signal that is not subjected to the filtering process for each filter input (the output of the register 62 for the horizontal one-dimensional low-pass filter and the output of the line memory 65 for the vertical one-dimensional filter). (The output of the weighted adding circuit 64 in the case of the horizontal one-dimensional low-pass filter, and the output of the weighted adding circuit 67 in the case of the vertical one-dimensional filter), and the absolute value of the difference is a threshold T.
If it is within the range, the signal subjected to the filtering process is selected as the output, and if it exceeds a certain threshold value T, the signal not subjected to the filtering process is selected as the output.

That is, in the direction perpendicular to the steep edge included in the image, the input and output of the filter are significantly different, so that the filter input is selected and the filter is turned off. On the other hand, in a direction parallel to the image edge or in a flat area not including a valid edge, a filter output is selected because the input and output of the filter take close values, and the filter is turned on in all directions.

As a result, a filter acts on a minute change portion such as a ringing area or a step between blocks existing around the edge of the image to reduce distortion, and at the image edge, the filter is turned off and the edge itself is preserved.

The functions described above are summarized as an adaptive low-pass filter. This filter portion is shown as a filter block 6 in FIG.

The output of the filter block 6 is connected to the image signal output terminal 7 and is also input to the frame memory 83. The frame memory 83 basically creates and outputs an image signal delayed by one frame, but it is possible to select whether to rewrite or hold the contents of the frame memory by a frame memory input write enable signal 85.

That is, the control circuit 82 instructs frame memory writing when the decoded image signal has changed from the previous frame and when the decoded image signal has not changed from the previous frame and is less than N frames, and When the number of frames becomes N or more, the operation is performed to hold the frame memory.

The control circuit 82 receives the parameter signal from the decoding device 4 via the parameter signal input terminal 8 and controls the control signal using the history. The parameter signal is, for example, as follows. The image signal is blocked and encoded by the encoding device 2. A signal indicating whether or not the signal has changed from the previous frame for each block is defined as a parameter signal. In this case, each block in the image passes the filter N times after the change occurs. This will be described with reference to FIG.

Reference numeral 101 in FIG. 2 shows a state in which whether or not an arbitrary block has changed is temporally changed in units of frames. In 101, state '1' is a parameter signal indicating that a block has changed in the frame, and state '0' is a parameter signal indicating that no change has occurred. Reference numeral 102 denotes a counter value for judging the history of the parameter signal. The counter value is set to an initial value N (N = 3 in the example of FIG. 3) in a frame having a change, and the decremented value is 0 in a frame having no change.
Then, 0 is continued until the next value N is set. This counter is easily realized as a function of the control circuit 82 in FIG.

The filter can be controlled using these two control signals.
FIG. 103 shows a filter input. First, the filter input selects the decoded image signal 'D' when the signal 101 indicates '1'.
When '0' is indicated, frame memory output 'F' is selected.

Next, the input to the frame memory 83 is in a write state “W” when the signal 102 is not “0”, and in a write stop state “H” when the signal 102 is “0”. FIG. 104 shows this state. As a result, the signal of FIG. 105 is obtained as a filter output.

That is, when a certain block changes, the filter is set to 1
The signal 'F1' that has passed through is output. When the change of the block stops, the signal that has passed through the filter in the frame memory passes through the filter again, and the signal 'F2' that has passed through the filter twice is output. When the state where there is no block change continues, the number of times of passing through the filter is sequentially increased and output as 'F3' and 'F4'. When it becomes 'F4', the frame memory is in the writing stop state, and thereafter, 'F4' continues. Next, when a change occurs, it returns to 'F1'. The figure shows the case where N = 3, and the maximum number of times of passing through the filter is N + 1 = 4. N =
When it is set to 0, the number of times of passing through the filter is 1, which corresponds to the conventional means.

FIG. 3 shows how the overall filter coefficient changes as the number of filter passes increases. FIG. 3 shows a coefficient distribution in the image space of the two-dimensional filter. However, the coefficient values indicate relative values, and are actually normalized so that the sum of the coefficients is 1. Symbols H and V indicate the number of passes of the horizontal low-pass filter and the vertical low-pass filter, respectively. For example, HHVV is a total filter coefficient when each filter is passed twice and the filters are turned on both horizontally and vertically, and a 5 × 5 two-dimensional low-pass filter is realized. Each time the number of passes increases by one, the order becomes 7
× 7,9 × 9, and a sufficient noise suppression effect can be realized.

At the edge of the signal, the horizontal or vertical filter is turned off and the filter operates only in the direction parallel to the edge, so that the image is not blurred. When the change of the image stops, the filter order gradually increases.
It is possible to prevent the distortion from continuing for a long time and causing a great disturbance visually.

Although one example of the adaptation has been described, a combination with another adaptation method is also possible. 2 shown in the explanation
The two parameters, the filter on / off threshold T and the counter initial value N, can be changed according to the strength of the target filter.

For example, if the coding noise increases, T is reduced to increase the number of filter-on states, and N is increased to increase the number of filter passage stages so that a large amount of noise can be removed.
If the coding noise decreases, T is increased, and N is decreased, so that blurring of the image is minimized. It is also possible to change the threshold value T in proportion to the step size for performing quantization.

Although the description has been made using the hardware circuit in the embodiment, the present invention can be similarly realized using a processor or the like used in the field of digital signal processing. In this case, the present invention relates to a post-filter processing algorithm using a low-pass filter having an adaptive feedback path using a frame memory.

(Effects of the Invention) As described above, according to the present invention, a decoded image signal is output after passing through a filter arranged in an output stage a plurality of times. It is possible to smooth and remove the step between blocks and the ringing of the image edge.

Further, since the number of times of passing through the filter and the on / off of the filter can be adapted, the image itself is not blurred. As a result, the image signal can be used as a post filter for removing a noise component from the image signal transmitted with high compression encoding.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a circuit configuration of an embodiment of the present invention, FIG. 2 is a time chart for explaining an operation of the embodiment of the present invention, and FIG. 3 is a post-filter coefficient realized by the present invention. FIG. 4 is a circuit example of a conventional post filter, and FIG. 5 is a circuit example of an improved conventional post filter. DESCRIPTION OF SYMBOLS 1 ... Image signal input terminal, 2 ... Encoding device, 3 ... Transmission path,
4 Decoding device, 5 Decoded image signal terminal, 6 Filter block, 7 Image signal output terminal, 61, 62, 63 Register for delaying line sequential scanning image signal by one pixel, 64, 67 ... Weighted Addition circuits 65, 66: a line memory for delaying an image signal by one scanning line; 68, an on / off selection circuit for a horizontal one-dimensional filter; 69, an on / off selection circuit for a vertical one-dimensional filter;
... a parameter signal input terminal, 82: a control circuit for determining the input condition of the filter and the input condition of the frame memory from the history of the parameter signal, 83 ... a frame memory, 84: a filter input selection circuit, 85 ... a frame memory input write enable signal, 86 ... Filter input selection signal.

──────────────────────────────────────────────────の Continued on the front page (51) Int.Cl. ⁶ Identification code FI H04N 1/41 H04N 1/41 B 5/21 5/21 Z 7/24 7/13 Z

Claims (1)

(57) [Claims] A signal processing method for removing a noise component from an image signal transmitted with a high-compression code, in particular, a terminal for inputting a decoded image signal, and wherein a value at each position of the decoded image is different from a value of a decoded image of a previous frame. In the configuration of a post-filter having a parameter signal input terminal for inputting a parameter signal indicating whether or not it has changed together with the decoded image signal, and an image signal output terminal for finally outputting a filter block output to the outside, A filter block that conditionally selects and inputs a signal or a frame memory output image signal, a frame memory that conditionally inputs and writes the output of the filter block, and an input condition of the filter block and an input of the frame memory based on the history of the parameter signal. A post comprising a control circuit for determining conditions. filter.