KR100715512B1 - Apparatus for image processing and method thereof - Google Patents

Abstract

An image processing apparatus and method thereof are disclosed. According to the present invention, a discrete cosine transformer generates a plurality of discrete cosine transform coefficients by performing a discrete cosine transform (DCT) on an input image, and outputs preprocessing data in which some of the discrete cosine transform coefficients are replaced with zeros by selecting a filter condition. And an inverse discrete cosine converter for generating a signal processed image by inverse discrete cosine transforming the preprocessed data. Conventional video / audio decoders with discrete cosine-to-inverse discrete cosine converters are capable of additional image signal processing implemented by separate image signal processing logic while minimizing structural changes.

Discrete Cosine Transform, Filter

Description

Apparatus for image processing and method

1 is a block diagram of a JPEG encoder / decoder, which is a representative image encoder / decoder.

2 is a block diagram of an image processing apparatus according to a preferred embodiment of the present invention.

FIG. 3 is a diagram illustrating region-specific characteristics of a discrete image obtained by transforming a cosine.

4A illustrates a filtering region of an 8 × 8 discrete cosine transform coefficient matrix in accordance with a preferred embodiment of the present invention.

4B illustrates the edge detection region of an 8 × 8 discrete cosine transform coefficient matrix in accordance with a preferred embodiment of the present invention.

Fig. 5 is a diagram showing a coefficient information table when the size of the input image is 8 × 8.

6 is a flowchart schematically showing an operation of an image processing apparatus according to a filter condition;

7 is a flowchart illustrating the operation of a data regulator.

The present invention relates to an image processing apparatus, and more particularly, to an image processing apparatus using characteristics of discrete cosine transform.

Discrete Cosine Transform is a method of dividing the original image into a predetermined size (4x4 to 32x32 pixels) and decomposing each divided region from the DC component to the highest frequency component. Most international standards bodies (ISO, ITU, IEC, FCC / HDTV) adopt discrete cosine transforms by default for video / audio encoding and decoding (e.g. JPEG, MPEG1, MPEG2, H.261, H.221, etc.) Doing.

1 is a diagram illustrating a configuration of a JPEG decoder, which is a representative image encoder-decoder, and the JPEG encoder 100 quantizes the discrete cosine converter 101 and the discrete cosine transform coefficients for converting original data. The quantizer 102 is composed of a zigzag scan 103 for zigzag scanning and outputting quantization coefficients, and an endoscope encoder 104 for encoding the output of the zigzag scan. The decoder 150 decodes an entropy code. An entropy decoder 151 for outputting quantization coefficients, an inverse zigzag scan 252 for reverse zigzag scanning quantization coefficients output from the entropy decoder, and an inverse quantizer 153 for restoring the quantization coefficients to discrete cosine transform coefficients. And an inverse discrete cosine transformer 154 that converts the discrete cosine transform coefficients to restore the original data. In the case of a video, a component (eg, intra frame prediction or inter frame prediction) for motion compensation is added. However, since the video / audio sub-decoder is intended for video / audio compression, the discrete and inverse discrete converters included in each are not used for purposes other than compression / restore.

Meanwhile, in addition to converting or reconstructing the input image into image data in JPEG format, an additional memory and logic are required for additional image signal processing such as filtering or edge extraction. Therefore, in the related art, since both the image encoder decoder logic and the separate image signal processing logic have to be included in one processor, there is a problem in that the configuration becomes complicated.

Accordingly, an object of the present invention to solve the above-described problem is to enable additional image signal processing such as filtering and edge detection by using the characteristics of the discrete cosine transform. In particular, it enables additional image signal processing implemented by separate image signal processing logic while minimizing the structure change of the existing video / audio decoder.

In order to achieve the above objects, according to an aspect of the present invention, a discrete cosine transformer for generating a plurality of discrete cosine transform coefficients by performing a discrete cosine transform (DCT) of the input image, among the discrete cosine transform coefficients by filter condition selection An image processing apparatus is provided that includes a data conditioner for outputting preprocessed data partially replaced by zero, and an inverse discrete cosine transformer for inverse discrete cosine transforming the preprocessed data to generate a signal processed image.

According to another aspect of the present invention, an encoder for discrete cosine transform coefficient (DCT) of an input image and outputting a plurality of discrete cosine transform coefficients and compressed data encoding the input image, respectively, the discrete cosine transform coefficients by filter condition selection There is provided an image processing apparatus including a data regulator for outputting preprocessed data, some of which is replaced with 0, and a decoder for selectively processing any one of the preprocessed data and the compressed data by a mode selection signal to generate an output image. .

According to another aspect of the present invention, the discrete cosine transform the input image, replace some of the plurality of discrete cosine transform coefficients corresponding to the input image with 0 according to the filter condition selection, and the partially replaced discrete cosine transform An image signal processing method is provided which includes outputting an image signal-processed according to the filter condition by inverse discrete cosine transforming coefficients.

The present invention can be applied to both a still picture decoding decoder (typically JPEG) and a video decoding decoder (typically MPEG-1, 2, 4) employing a discrete cosine transform, but the present invention clearly explains the gist of the invention. In order to apply the present invention to a still image decoding apparatus, an embodiment of generating a still image (before processing) -still image (after processing) is mainly described. However, those skilled in the art may apply the video decoding apparatus to a video decoding unit, as well as a still image (before processing) -still image (after processing), as well as a moving picture (before processing) -video (after processing). Modifications such as can be easily implemented. Hereinafter, with reference to the accompanying drawings will be described in detail a preferred embodiment of the present invention.

2 is a block diagram of an image processing apparatus according to a preferred embodiment of the present invention.

The image processing apparatus includes a discrete cosine transformer 201, a data controller 260, an inverse discrete cosine transformer 254, and includes a quantizer 202, a zigzag scan 203, an entropy encoder 204, an entropy decoder ( 251, a reverse zigzag scan 252, a dequantizer 253, and a multiplexer 270. In addition, the image processing apparatus may further include an image transform unit for dividing the original image into a downsampling transform, a color space transform, and a minimum coded unit (MCU) in front of the discrete cosine transformer 201. have. However, according to the JPEG standard, downsampling conversion and color space conversion are optional items, and thus, in this specification, a method of converting and processing the original image into MCU blocks, which are one data unit, will be described. Of course, in the case of MPEG, the input image may be a macro block 16X16.

The discrete cosine converter 201 transforms the input image (typically an 8 × 8 pixel block) to produce a discrete cosine transform coefficient. In the pixel area, image information is widely spread over an entire portion of the image. The discrete cosine transform of the information in the pixel region is converted into the spatial frequency domain, which causes the image information to converge toward the DC (average of the image in the pixel region) and the low frequency. In particular, human vision is sensitive to DC values and low frequency components in the spatial frequency domain, but insensitive to high frequency components. FIG. 3 is a diagram illustrating the region-specific characteristics of the discrete cosine transformed base image (size NxN). The first discrete cosine transform coefficients correspond to an average value of the input image, and the coefficients correspond to low to high frequency (AC). do. In the region where discrete cosine transform coefficients of the AC component are distributed, as shown in FIG. 3, the coefficients corresponding to the low frequency are concentrated in the upper left, and the coefficients corresponding to the mid frequency are concentrated in the upper right and lower left. It can be seen that the coefficients corresponding to the frequency are concentrated on the lower right side and the right side and the lower side. In addition, the coefficients corresponding to the vertical edge of the input image are mainly concentrated in the upper right, the coefficients corresponding to the horizontal edge are mainly concentrated in the lower left, and the coefficients corresponding to the entire edge are lower right. It can be seen that the coefficients concentrated at and corresponding to the diagonal edges are concentrated at other regions.

Referring back to FIG. 2, the data regulator 260 is coupled to the discrete cosine transformer 201 and the inverse discrete cosine transformer 254, respectively, and converts some of the discrete cosine transform coefficients output from the discrete cosine transformer 201 to zero. Substitute the preprocessing data. The data conditioner 260 replaces the specific discrete cosine coefficients with zeros according to a preset filter condition selection, which filter the low pass filtering, the midrange filtering, the high pass filtering, the diagonal edge detection, the vertical edge detection, the horizontal edge detection, the full edge. Either detection. Coefficient information (e.g., information about which coefficient to convert to 0) about coefficients to be replaced with zero according to each filter condition selection may be stored in advance in the data controller 260 or may be input externally. have. When coefficient information is externally input, the filter condition selection itself may be coefficient information. The coefficient information input from the outside is stored in a register inside the data controller 260 or in a separate memory (not shown) located inside the image processing apparatus. The processing by the data controller 260 will be described in detail with reference to FIGS. 4 to 6.

The inverse discrete cosine converter 254 inversely discrete-converts the preprocessed data generated by the data controller 260 by replacing some coefficients by zero by filter condition selection to generate a signal processed image. Since the coefficients located outside the region corresponding to the filter condition selection are replaced with zeros in the preprocessing data, the input image itself is treated as an image according to the filter condition selection (for example, an image including only low frequency components). Thus, when the preprocessed data is inversely transformed by the inverse discrete cosine transformer 254, an image corresponding to the filter condition selection is finally generated. In particular, since the input image itself is directly signal processed without undergoing quantization-dequantization, there is no information loss.

In addition to the image signal processing, the image processing apparatus described above may further include a component for operating as the decoders 200 and 250 for converting the input image into JPEG format data or restoring JPEG format data to the input image.

The encoder 200 further includes a quantizer 202, a zigzag scan 203, and an entropy encoder 204 in addition to the discrete cosine converter 201, and the quantizer 202 is a discrete output from the discrete cosine converter 201. Quantize the cosine transform coefficients to produce quantization coefficients. The generated quantization coefficients are scanned by zigzag scan 203 so that the variance is reduced in algebraic form as the frequency is generally higher. The zigzag scanned quantization coefficients are encoded by the entropy encoder 204 and finally converted into JPEG format data. In MPEG, an alternate scan for interlaced scan may be applied in addition to a zigzag scan for progressive scan or non-interlaced scan. Typical entropy encodings include Huffman encoding, run-length encoding, and arithmetic compression.

The decoder 250 reversely performs a function corresponding to each component of the encoder 200 in order to restore JPEG format data to an input image. The multiplexer 270 controls the input of the inverse discrete cosine converter 254 according to an externally input mode selection signal. That is, by the mode selection signal, only one of the discrete cosine transform coefficients output from the inverse quantizer 253 and the preprocessed data output from the data controller 260 may be input to the inverse discrete cosine converter 254.

4A and 4B illustrate a filtering area and an edge detection area of an 8 × 8 discrete cosine transform coefficient matrix according to a preferred embodiment of the present invention. Here, the regions according to the filter conditions illustrated in FIGS. 4A and 4B are selected when the size of the input image is 8x8, and may be changed accordingly when the size of the input image changes.

4A illustrates discrete cosine transform coefficients of an input image composed of 8x8 pixels by using the region-specific characteristics of the discrete cosine transformed base image shown in FIG. 3. And divided into regions (a1, a2, a3, a4) necessary for full edge detection, and indicate indexes of coefficients belonging to each region. The coefficients corresponding to the diagonal edge a1 are 2, 3, 4, 9, 10, 11, 12, 17, 18, 19, 20, 21, 25, 26, 27, 28, 29, 30, 35, 36, Coefficients 37, 38, 44, 45, coefficients corresponding to the horizontal edge a2 are coefficients 33, 34, 41, 42, 43, 49, 50, 51, 52, 57, 58, 59, 60, The coefficients corresponding to the vertical edge a4 are the 5th, 6th, 7, 8th, 13th, 14th, 15th, 16th, 22th, 223th, 24th, 31st, 32nd coefficients, and the coefficients corresponding to the entire edge a3 are 39 , 49, 46, 47, 48, 53, 54, 55, 56, 61, 62, 63, 64th coefficient. In edge detection, the DC coefficient does not use the first coefficient.

FIG. 4B illustrates regions required for low-pass frequency filtering, mid-frequency filtering, and high-frequency frequency filtering of the discrete cosine transform coefficients of an input image composed of 8x8 pixels by using the region-specific characteristics of the discrete cosine-converted base image shown in FIG. 3. b1, b2, and b3), and the indexes of the coefficients belonging to each area are indicated. The coefficients corresponding to the low frequency b1 are 1, 2, 3, 9, 10, 11, 17, 18th coefficients, and the coefficients corresponding to the mid frequency b2 are 4, 5, 6, 7, 8, 12 , 13,14, 15, 19, 20, 21, 22, 23, 25, 26, 27, 28, 29, 30, 31, 33, 34, 35, 36, 37, 41, 42, 43, 44, 45 , 49, 50, 51, 57th coefficients, the coefficients corresponding to the high frequency (b3) are 16, 24, 32, 38, 39, 40, 46, 47, 48, 52, 53, 54, 55, 56, 58, 59, 60, 61, 62, 63, 64th coefficient. FIG. 5 shows coefficient information according to filter conditions determined by FIGS. 4A and 4B in a range form. Here, CNT represents a counter value, and the counter value is equal to the index of the coefficient.

FIG. 6 is a flowchart schematically illustrating an operation of processing an image signal according to a filter condition by an image processing apparatus implemented as an encoder. In operation 300, an image processing apparatus is initialized and a mode selection signal is input. The mode selection signal includes a normal mode for the image processing apparatus to perform a JPEG (or MPEG) compression / restore operation and an image signal processing (ISP) mode for performing frequency filtering / edge detection. Image signal processing modes include frequency filtering (low frequency filtering, mid frequency frequency filtering, high frequency frequency filtering) and edge detection (diagonal edge detection, vertical edge detection, horizontal edge detection, full edge detection) as filter conditions. Of course, the image signal processing mode signal and the filter condition may be divided signals.

In step 305, it is determined whether the mode selection signal is an image signal processing mode. If the determination result is not in the image signal processing mode, the input image is converted into JPEG format data or the JPEG format data is restored to the same output image as the input image.

If it is determined in step 315 that the image signal processing mode is determined, it is determined whether the filter condition is edge detection. If the determination result is low frequency filtering, in step 325, coefficients belonging to regions other than the b1 region are replaced with zeros. If it is not low frequency filtering in step 330, it is determined whether it is mid-frequency frequency filtering in step 330, and according to the determination result, the coefficients belonging to areas other than b2 are replaced with 0 (step 335), or the coefficients belonging to areas other than b3 are replaced. Replace with 0 (steps 340 and 345).

If it is determined in step 350 that the filter condition is edge detection, it is determined whether the diagonal edge is detected. If the result of the determination is diagonal edge detection, the coefficients belonging to areas other than the a1 area are replaced with 0 (step 355). In operation 360, if the result of the determination is not a diagonal edge detection, it is determined whether the detection is a horizontal edge detection. If the determination is a horizontal edge detection, the coefficients belonging to a region other than the a2 region are replaced with 0 (step 365). In operation 370, it is determined whether the detection is the horizontal edge detection or the vertical edge detection, and according to the determination result, the coefficients belonging to the area other than the a4 area are replaced with 0, or the coefficients belonging to the area other than the a3 area are replaced with 0.

The preprocessed data with coefficients replaced by zero is input to an inverse discrete cosine converter and converted into an output image which is a signal processed image.

7 is a flowchart illustrating the operation of the data controller. In step 405, the data controller selects coefficient information corresponding to the filter condition. Coefficient information (see FIG. 5) corresponding to each filter condition is stored in a memory located inside the data controller or inside the image processing apparatus or input externally with a mode selection signal.

In step 410, after resetting the counter to count the number of coefficients to be sequentially input from the discrete cosine converter is incremented by one. When the input image is an 8x8 pixel block, a total of 64 coefficients are input.

In step 415, the data adjuster determines whether the i th coefficient is included in the coefficient information.

If the determination result is not included in step 420, the i th coefficient is output to the inverse discrete cosine converter. If it is determined in step 425, the i-th coefficient is replaced with 0 and output to the inverse discrete cosine converter.

In operation 430, it is determined whether the last pixel of the input image is referred to by using the counter value. If the result of the determination is not the last pixel, the counter value is increased by 1 in step 435.

If it is the last pixel in step 440, the counter is reset to process the next input image.

In step 445, it is determined whether the image is the last input image. If it is not the last input image, the process proceeds to step 435 where the counter value is increased by one.

As described above, the present invention enables image signal processing using only the discrete cosine converter-inverse discrete cosine converter. In particular, the conventional video / audio encoding / decoder having a discrete cosine converter to an inverse discrete cosine converter can process additional image signals implemented by separate image signal processing logic while minimizing structural changes.

In addition, since the input image is immediately converted without input through the quantizer-dequantizer, no loss of the input image occurs. Therefore, the quality of the output image is superior to that of the prior art which converts the input image into a still image (JPEG) or moving image (MPEG) format and then performs image signal processing.

Although the above has been described with reference to a preferred embodiment of the present invention, those skilled in the art to which the present invention pertains without departing from the spirit and scope of the present invention as set forth in the claims below It will be appreciated that modifications and variations can be made.

Claims (10)

A discrete cosine transformer for generating a plurality of discrete cosine transform coefficients by performing discrete cosine transform (DCT) on the input image; A data conditioner for outputting preprocessing data in which some of the discrete cosine transform coefficients are replaced with zeros by filter condition selection; And And an inverse discrete cosine converter for generating the signal-processed image by inverse discrete cosine transforming the preprocessed data. The image processing apparatus of claim 1, wherein the input image is a minimum coded unit (MCU). The image processing apparatus of claim 1, wherein the filter condition selection is any one of low pass filtering, band call filtering, and high pass filtering. The method of claim 3, wherein the discrete cosine transform coefficients are arranged in the form of an NxN matrix, and are divided into a low frequency band, a mid frequency band, and a high frequency band according to an arrangement position. And the data controller replaces discrete cosine transform coefficients, which are not included in a frequency band selected according to the filter condition selection, with zero. The image processing apparatus according to claim 1, wherein the filter condition selection is any one of horizontal edge detection, vertical edge detection, diagonal edge detection, and full edge detection. The method of claim 5, wherein the discrete cosine transform coefficients are arranged in the form of an NxN matrix, and are divided into DC coefficients, horizontal edge regions, vertical edge regions, diagonal edge regions, and all edge regions according to the arrangement position. And the data controller replaces discrete cosine transform coefficients, which are not included in the selected region, with zero according to the filter condition selection. An encoder for performing discrete cosine transform (DCT) on the input image to output a plurality of discrete cosine transform coefficients and compressed data encoding the input image; A data conditioner for outputting preprocessing data in which some of the discrete cosine transform coefficients are replaced with zeros by filter condition selection; And And a decoder configured to selectively process any one of the preprocessed data and the compressed data by a mode selection signal to generate an output image. The method of claim 7, wherein the decoder An entropy decoder for decoding the compressed data and outputting a plurality of quantization coefficients; An inverse quantizer for converting the quantization coefficients into discrete cosine transform coefficients; A multiplexer for selecting any one of the discrete cosine transform coefficients output from the preprocessing data and the inverse quantizer by the mode selection signal; And And an inverse discrete cosine converter for generating an output image by inverse discrete cosine transforming the output of the multiplexer. In the image signal processing method of the video encoder, Discrete cosine transforming the input image; Replacing some of the plurality of discrete cosine transform coefficients corresponding to the input image with zeros according to a filter condition selection; And And inverse discrete cosine transforming the partially substituted cosine transform coefficients to output an image signaled according to the filter condition. The method of claim 9, wherein, in response to the filter condition selection, replacing some of the plurality of discrete cosine transform coefficients corresponding to the input image with 0 includes: Selecting coefficient information corresponding to the filter condition, wherein the coefficient information is information for identifying a coefficient included in an area corresponding to the filter condition; Bypassing coefficients included in the coefficient information; And And replacing coefficients not included in the coefficient information with zeros.

Images