JPH04220084A - Picture coding system - Google Patents

Abstract

PURPOSE:To apply compression processing to a gradation area extracted from an input picture by means of ADCT and to apply coding to a non-gradation area by using an arithmetic code expanded in multi-level processing. CONSTITUTION:A color conversion section 2 converts an input picture signal into a luminance component.and a color difference component. Only the luminance component is inputted to a gradation area extraction section 3, in which a gradation area is extracted from a picture data, a coding section 4 uses an ADCT to apply compression processing to the extracted gradation area. After areas except the gradation area in the input picture are subject to color limit processing, the gradation area is set to 0 by a flag processing section 6 and the data of the areas subjected to color limit processing and the gradation area set to 0 are coded at a coding section 7 by using an arithmetic code expanded to a multi-level.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image encoding system for an image recording device such as an electronic file.

0002

2. Description of the Related Art In recent years, with the development of multimedia, an increasing number of office documents are expressed in color images. Such color document images have a much larger amount of data than conventional monochrome binary documents, and even when storing color document images using a large-capacity storage medium such as an optical disk, the amount of data is extremely large. Compression processing is required to reduce this.

By the way, in general, color document images include:
Binary images such as character areas and gradation areas such as pictures coexist, and since these areas have different frequency characteristics, it is not efficient to process them using the same encoding process. compresses color document data by using a compression method suitable for each area.

[0004] That is, one-dimensional encoding (Modified Hu
ffman encoding system, MH) or two-dimensional sequential processing encoding (ModifiedMR encoding system, MMR)
For general binary images, PRES
+ encoded with arithmetic code, A for natural still image areas
DCT (Adaptive Discrete Cosi
Compression is performed using ne Transformer (applied discrete cosine transform).

[0005] As described above, in general, a reversible encoding method is used for an image having many binary edge components, and an irreversible compression method is used for an image having many tone components. Because of the differences in the characteristics of such encoding methods, in so-called multimedia images in which images with different characteristics coexist, the regions are divided and adaptive encoding processing is performed for each region. For example, an encoding method has been proposed in which a character region is extracted as a binary region, the character region is encoded by MMR, and the remaining region is compressed by ADCT (1990 National Conference of the Institute of Image Electronics Engineers). Proceedings 31, P131-136).

[0006]

[Problems to be Solved by the Invention] The region separation method in the above-mentioned encoding system separates regions only for text regions, so in the case of complex images such as catalogs, it is difficult to separate text and pictures. The problem was that it was difficult. Conventional region separation methods to solve this problem include, for example, an algorithm that focuses on edge elements to determine block attributes, focuses on background density to determine block attributes, and then integrates blocks to extract light and shade regions. (IEICE Journal '87/1 Vol. J7
0-BNo. 1pp88-95). However, since this algorithm uses background density, there is a problem in that it is difficult to determine the background density when multiple color background regions exist in an image.

An object of the present invention is to extract tone regions from an input image, compress the tone regions by ADCT, and compress the non-gradation regions by using arithmetic codes expanded to multi-values. The object of the present invention is to provide an image encoding method for encoding.

[0008]

Means for Solving the Problems In order to achieve the above object, the invention according to claim 1 provides an image encoding method for encoding a color image, which includes means for extracting a gradation region from an input image; a first encoding means for encoding a gradation region; a means for converting a region of the input image excluding the gradation region into a limited color; and setting the gradation region to a predetermined attribute value. and second means for encoding the limited color area and the gradation area set to the attribute value.

In the invention according to claim 2, the means for extracting the gradation area determines the gradation of the area by referring to the edge component in the local area and the normal vector direction of the approximate plane of the grayscale image. It is characterized by

The invention according to claim 3 is characterized in that the size of the pixel area necessary for determining the approximate plane is dynamically changed so as not to include edge pixels. The invention according to claim 4 is characterized in that a local area in which the number of edge pixels is less than or equal to a predetermined threshold and the normal vector is inclined at a predetermined angle or more with respect to the Z axis is determined to be a gradation area. There is.

[0011]

[Operation] According to the image encoding method of the present invention, the input image signal is converted into a luminance component and a color difference component by the color conversion section, and only the luminance component is inputted to the gradation region extraction section, and the image data is A gradation region is extracted from the gradation region, and the encoding unit performs compression processing on the extracted gradation region by ADCT. After converting the input image into a limited color for the area excluding the gradation area, the gradation area is set to a value of 0 in the flag processing section, and the limited color area and the gradation set to the value 0 are set. The region is encoded using an arithmetic code expanded to multi-value in the encoding section. Therefore, according to the present invention, the gradation area is extracted from a multi-level color original, and the encoding method is divided into the gradation area and the non-gradation area, so that color characters whose shapes cannot be defined, etc. Even when a color background is included, non-gradation areas can be compressed by the same encoding process without recognizing the shapes of those areas, and overhead due to area information does not increase.

Furthermore, since the local gradation of an image is determined based on edge information and the slope of the normal vector, it is possible to extract gradation areas even in documents where the background color of the image cannot be determined. The same process can be used to determine the gradation of an image expressed in pseudo-halftones, such as a halftone image. Furthermore, since the determination results of the local processing are checked by the global processing, it is possible to extract even areas where the gradation gradually changes over a wide range.

Since the size of the mask is adaptively switched so as not to include edge pixels, it is possible to reduce the estimation error of plane approximation around the edges, and also to reduce the number of edge pixels and normal vectors. Since a block that is tilted by a predetermined angle or more with respect to the Z axis is determined to be a gradation area, it is possible to determine the gradation of a local area with high accuracy.

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described in detail below with reference to the drawings. For example, when encoding a color document image with a complex structure such as a catalogue, the fewer encoding methods are used, the simpler the processing will be, and the less header information will be required when storing or transmitting. In the present invention, a color document image is separated into two types of areas, a gradation area and a non-gradation area, focusing on the gradation of the image.
A method is used to encode each area.

FIG. 1 is a block diagram of an image encoding apparatus according to an embodiment of the present invention. An input image is read by a multi-gradation image input section 1, and the read image data is decomposed into R, G, and B signals (each color has 256 gradations) and input to a color conversion section 2. In the color conversion section 2, R, G
, the luminance component Y and the color difference components Cr, C obtained by linearly converting the B signal.
Convert to b. Of these, only the luminance component Y is input to the gradation region extraction section 3, which extracts the gradation region from the image data as described later. In this way, since the present invention focuses on the gradation of an image, it is possible to extract a gradation region from only the luminance component. Next, in the gradation region encoding unit 4, the extracted gradation region is encoded block by block sequentially from the upper left to the lower right like raster scanning using the ADCT method. Note that ADC, which is a gradation domain encoding method,
In the T method, 8 pixels x 8 pixels are encoded as one block, but the color difference components Cr and Cb are thinned out to 1/2 in the horizontal direction (or vertical direction), so the unit of encoding processing block is 16 The number of pixels is 16 pixels. Therefore, one block has 16 pixels x 16 pixels in the extraction unit of the gradation region.

On the other hand, after the encoding process of the gradation area, the encoding process is performed on the non-gradation area, but as pre-processing,
Performs color limiting processing for non-gradation areas and flag processing for gradation areas in the original image. That is, in the state of the original image, density values are not stable even within the same character or between adjacent pixels on the same background due to the influence of noise, so it is necessary to perform limited colorization of the image. The limited color processing unit 5 takes a density histogram of the entire non-gradation image and maps it with 255 representative colors.

By the way, since the correlation between pixels in a non-gradation area is very strong, higher compression is generally possible than in a gradation area, and even if the entire image is encoded, the amount of encoded data is small. Therefore, in the present invention, the entire original image is encoded to suppress the increase in header information. At this time, in order to identify the previously extracted gradation area, the flag processing unit 6 replaces the gradation area part with a flag with an attribute value of 0 (since the value 0 is used for the flag, the representative color described above is (set in the range of values 1-255). For the non-gradation image containing flag information generated in this way, the non-gradation encoding processing unit 7 applies an arithmetic code expanded to a multi-value (this code is described in, for example, Journal of the Institute of Image Electronics Engineers, 12, 3pp 219-226 1983). FIG. 2(a) shows an original image including a gradation area such as a photograph, and FIG. 2(b) shows an image in which the photograph part is replaced with flag 0 and the entire image is encoded as a non-gradation image.

FIG. 3 is a flowchart showing the procedure of the gradation area extraction process in the gradation area extraction section 3. below,
The gradation region extraction process according to the present invention will be explained in detail.

Edge extraction (step 31): Edges are extracted using the Sobel operator. By setting the threshold value for determining whether or not an edge is an edge to a value of 250, edges of solid characters or background are extracted. Furthermore, in the case of a pattern expressed by halftone dots such as in a printed matter, when input using a scanner, density changes do not appear very large, so edges hardly occur. Therefore, even in a printed matter, if there are many edge pixels, there is a high possibility that it is a non-gradation area such as a character area.

Calculation of normal vector (step 32); Normal vectors are calculated for pixels other than the edge pixels extracted in the above step. That is, considering a grayscale image based on luminance component values as a three-dimensional figure, a straight line perpendicular to the tangential plane of the point P for which the normal is to be found is called the normal to the point P, and its direction vector is called the normal vector. The normal vector in a grayscale image is approximated by considering a small region surrounding a pixel of interest as a surface element, and by regarding this small region as a plane. Then, the normal vector of this approximated plane is taken as the normal vector of that plane element (FIG. 4). Specifically, the plane equation is a
With x+by+cz=1, approximation is performed using the least squares method using the spatial coordinates of pixels in the small area. Note that the estimation error increases when the approximate plane passes near the origin (0, 0, 0), so the spatial coordinate values are: x is the horizontal address of the pixel, y is the pixel's vertical address, and z is the luminance component value + 256. to move the approximate plane horizontally so that it does not pass through the origin.

Furthermore, the size of the pixel area (referred to as mask size) necessary for finding an approximate plane is determined by determining a maximum mask size and a minimum mask size, and dynamically changing the size between them. That is, as shown in FIGS. 5A and 5B, in order to improve calculation accuracy, the maximum mask size is used for areas other than edges, and the minimum mask size is used near edges. If an edge is included even if the mask size is set to the minimum size, the pixel is treated as a pixel for which the normal vector cannot be calculated.

Determination of gradation block (step 33): With reference to the edges and normal vectors obtained in the above step, the gradation area is determined for each block. In other words, if the percentage of edge pixels is above a predetermined threshold, there is a high possibility that it is a character edge, and if the normal vector is uniformly parallel to the Z axis, it is determined that it is a locally non-gradation area. Therefore,
If the proportion of edge pixels in a block is, for example, 6% or less (that is, 15 pixels, based on actual measurements), and the spatial angle between the normal vector in that block and the Z axis is tilted by 5 degrees or more, then Determine the block as a gradation area.

Region division (step 34): In the above step, gradation regions are extracted on a block-by-block basis, but a color document image may have a gradual gradation over a wide range. Therefore, in the present invention, the entire image is divided into regions, and the gradation of each divided region is determined. In other words, the region division is
The percentage of pixels for which normal vectors can be calculated is 6 within one block.
0% or more, and all normal vectors are approximately parallel to the Z axis (within 5 degrees in spatial angle). Then, four neighboring target blocks are merged as the same area. By checking the gradation of each group divided in this way again, the attribute of the region is determined globally.

Gradation determination for each area (step 35): Take the moving average of the density histogram for each area grouped in the above step, and if the maximum value of the moving average exceeds 90% of the total number of pixels, The entire area of the group is determined to be a single density area, that is, a non-gradation area, and its average density is set as the density value of all pixel values. In the case of 50% to 90%, the non-gradation area and the gradation area are mixed, and the area within the peak density range for each block is determined to be the non-gradation area. If it is 50% or less, it is assumed that gradation exists overall, and the entire area of the group is determined to be a gradation area. As a result, a global gradation region was determined.

Integration of regions (step 36); finally,
The areas are integrated so that the gradation area and the non-gradation area are not segmented too much. That is, the integration process is performed by changing the non-gradation area surrounded by the gradation areas in the four neighborhoods to the gradation area. As described above, in the present invention, since the gradation of an image is determined based on the slope of the normal vector, it is possible to suppress erroneous determination in halftone dot areas.

[0026]

[Effects of the Invention] As explained above, according to the present invention, the following effects can be achieved. That is, according to the invention set forth in claim 1, the gradation area is extracted from a multivalued color original, and the encoding method is divided into the gradation area and the non-gradation area, so that the shape cannot be limited. Even if the image contains colored characters or colored backgrounds, it is possible to compress the non-tone areas using the same encoding process without recognizing the shape of those areas, thus increasing the overhead due to area information. I won't let you.

According to the second aspect of the invention, since the local gradation of an image is determined based on the edge information and the slope of the normal vector, the gradation can be determined even in an original where the background color of the image cannot be determined. Regions can be extracted, and the gradation of an image can be determined by the same processing even for an image expressed in pseudo halftones, such as a halftone image. Furthermore, since the determination results of the local processing are checked by the global processing, it is possible to extract even areas where the gradation gradually changes over a wide range.

According to the third aspect of the invention, since the size of the mask is adaptively changed so as not to include edge pixels, it is possible to reduce estimation errors in plane approximation around edges.

According to the fourth aspect of the invention, a block having a small number of edge pixels and whose normal vector is tilted at a predetermined angle or more with respect to the Z axis is determined to be a gradation area, so that high precision can be achieved. It is possible to determine the gradation of a local area.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an image encoding device according to an embodiment of the present invention.

FIG. 2(a) shows an original image including a gradation area such as a photograph. (b) is an image in which the photographic part is replaced with flag 0 and the entire image is encoded as a non-gradation image.

FIG. 3 is a flowchart illustrating a gradation region extraction processing procedure in a gradation region extraction section of the present invention.

FIG. 4 is a diagram showing normal vectors of a plane approximated to a plane.

FIG. 5(a) is a plan view of a region for explaining maximum mask size and minimum mask size. (b) is a concentration cross-sectional view of a region for explaining the maximum mask size and the minimum mask size.

[Explanation of symbols]

1 Image input section 2 Color conversion section 3 Gradation area extraction part 4 Gradation domain encoding unit 5 Limited colorization processing section 6 Flag processing section 7 Non-gradation area encoding unit

Claims (4)

[Claims] 1. In an image encoding method for encoding a color image, means for extracting a gradation region from an input image;
a first encoding means for encoding the gradation area; a means for converting an area of the input image excluding the gradation area into limited colors; and a first encoding means for encoding the gradation area; An image encoding method comprising: a means for setting; and a second means for encoding the limited color area and the gradation area set to the attribute value. 2. The means for extracting the gradation region determines the gradation of the region by referring to the edge component in the local region and the normal vector direction of the approximate plane of the grayscale image. Image encoding method according to item 1. 3. The image encoding method according to claim 2, wherein the size of the pixel area required for determining the approximate plane is dynamically changed so as not to include edge pixels. 4. A local area in which the number of edge pixels is less than or equal to a predetermined threshold and the normal vector is tilted at a predetermined angle or more with respect to the Z axis is determined to be a gradation area. The image encoding method described in 2.