JP2012044502A - Medical color image compression method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an encoding technique having higher efficiency of compression in encoding a first difference of each color element of RGB.SOLUTION: A medical color image compression method for compression encoding of a digital color image taken for medical use comprises a first step of separating each pixel data into RGB, a second step of calculating difference data between adjacent pixels with respect to each of RGB, and a third step of selecting an optional element from among RGB, and comparing and encoding the difference component of the element and the difference components of the other two elements.

Description

  The present invention relates to coding for efficiently compressing color images widely used in the ophthalmic field. As for coding of image compression, there are reversible compression in which data is restored to the original at the time of decoding and irreversible compression in which data is not restored.

  The eye surface image is photographed by illumination with white light, or a state stained with a staining solution is photographed, or is observed through an optical filter for recording a specific wavelength with respect to the stained image. In order to observe the surface of the eye in detail, at least the cornea portion is captured in one image, and in order to obtain a sufficient resolution, generally an image of at least about 480 pixels and 640 pixels is taken. These are recorded as data files and recorded on a computer recording medium for digital interpolation, or are captured and recorded in a database structure. In the case of a color image, one pixel is divided into three primary colors of RGB, each having an 8-bit expression capability, and is generally expressed by 24 bits per pixel. An image with a larger size is used to visually recognize a blood vessel.

  For these large-format image recordings, a general JPEG format or the like is used as a digital recording of a computer, but these are irreversible compressions, and when reproduced on a monitor, the expression at the time of shooting is completely reproduced. I can't do it. The effect of this imperfection is conspicuous in the part where the color change is relatively large, and in an image having a shape on a thin line segment like a blood vessel spreading on a uniform background, it is due to restoration that leads to misperception Noise is generated.

  For lossless compression of general digital images, there is a run length method. In this case, when the same luminance appears continuously, encoding is performed with the number of appearances of the luminance value, and the compression efficiency increases as the number of places where the image is expressed with the same luminance. However, in a general natural landscape or an image of a living body, it is very rare that a region expressed by one luminance is continuous as the bit length assigned to one pixel becomes longer. It is not effective for the image.

  In addition, the compression by the difference method is a method of taking the difference from the neighboring pixels and reducing the number of bits by utilizing the fact that the neighboring pixels are not greatly different in an image of a general natural landscape or a living body. . There is also a method of calculating and compressing the difference between the difference values. The values taken by general pixels are considered to be 256 gradations from 0 to 255 in a state of being separated into black and white or RGB. Even if the difference data is used here, it becomes data with positive and negative signs, and the compression effect is not always obtained depending on how to handle the signed difference data.

  In order to solve the above problem, there is a patent application related to a data compression program disclosed in Patent Document 1. In this method, a difference is taken for each color element obtained by color-separating an image to reduce the amount of variation between adjacent pixels, and then encoding is performed. Further, the second difference is obtained in relation to the first difference for each color element and the other color elements. In encoding the difference value, an entropy encoding method is used. Therefore, regarding the encoding of the difference value, the portion showing novelty is limited.

  Further, there is an application related to a color image compression method disclosed in Patent Document 2. In this method, the first and second difference values are obtained for each color element divided into RGB and run-length compression is performed, and no novelty is found in the encoded portion.

JP 2007-174464 A JP 2003-299120 A

  Although the conventional primary difference and secondary difference are used to encode data, since an efficient method has not been applied, the present invention encodes the first difference for each color element of RGB. An object of the present invention is to provide an encoding technique with higher compression efficiency.

  In order to solve the above-described problems, the present invention is characterized by having the following configuration.

(1) In a medical color image compression method for compressing and encoding a digital color image taken for medical use, a first step of separating each pixel data into RGB, and calculating difference data between adjacent pixels for each of RGB And a third step of selecting an arbitrary element from RGB and comparing and encoding the difference component and the difference component of the remaining two elements. To do.
(2) In the medical color image compression method according to (1), in the second step, the difference data is all represented by 4 bits, and the numerical value is composed of 1 bit of code and 3 bits of data, or It consists of a marker that distributes information to the next 4 bits.

  INDUSTRIAL APPLICABILITY The present invention is used for image compression when a color image obtained by color photographing of the eye surface is digitally filed and stored on a computer. The image format provided for processing is such that each pixel component is separated for each color element that has not been subjected to compression processing, and its capacity is reduced by compression, and data is reversibly obtained by decoding processing. When restored, it is possible to reproduce a color expression that is completely the same as before compression. The compression rate differs for each image used for processing.

FIG. 3 is an explanatory diagram illustrating a flowchart of the first embodiment.

  Hereinafter, description will be made based on the flowchart shown in FIG. It is assumed that the target color image is expressed in RGB format. The resolution is expressed by 8 bits (0 to 255 gradations) for each of RGB. Even if it is a color image of a format other than this, it is generally easy to convert it to RGB. It is assumed that the image is composed of vertical H pixels and horizontal W pixels.

  A plane based on an arbitrary plane in each of the RGB planes. Here, for explanation, R will be described as a base plane. As shown in 1 of FIG. 1, first, the data of the R plane is sequentially scanned to create a difference table with adjacent data. Although the scanning direction is not defined, scanning is generally performed from the upper left to the right end of the image, and when the first line is completed, calculation is sequentially performed from the left end to the right end of the second line. Here, an apparent discontinuous point is generated by moving from the first line to the second line, but since it is not so many in the whole image, it continues as it is. It is also possible to change the scanning method to move from the right end of the first line to the right end of the second line as it is, and to scan leftward in the second line. In this description, it is assumed that each line is repeatedly scanned from the left end to the right end. The discontinuous points are handled as a continuous data string up to the lower right end of the image without particular mention. Further, in a natural image, the amount of calculation is reduced by processing only the primary difference by utilizing the fact that an extremely large change does not occur between adjacent pixels.

  A difference table is similarly created for G and B. Further, as shown in 2 of FIG. 1, a bit table that gives “1” when the difference values of G and B are the same as R and “0” when they are different is created separately from this difference table. This bit table has a table of H × B / 8 bytes for each of G and B.

  First, the R difference table is encoded. As indicated by 3 in FIG. 1, the difference values are encoded every 4 bits, and are sequentially organized as 8-bit data. Here, P is an arbitrary pixel data, and D is a difference value from the previous pixel. However, in the case of the first pixel, the value of P is encoded.

  First, when | D | ≦ 7. The encoded data is represented by 1 byte. If 0 ≦ D, the first byte is “D”. If D <0, the first byte is “8-D”.

  Next, when 8 ≦ | D | ≦ 119. The encoded data is expressed by 12 bits (3 nibbles). The first nibble is set to “08h”. The second nibble is set according to D as follows. “0010b” (8 ≦ D ≦ 23), “0100b” (24 ≦ D ≦ 39), “0110b” (40 ≦ D ≦ 55), “1000b” (56 ≦ D ≦ 71), “1010b” (72 ≦ D ≦ 87), “1100b” (88 ≦ D ≦ 103), “1110b” (104 ≦ D ≦ 119), “0011b” (−23 ≦ D ≦ −8), “0101b” (−39 ≦ D ≦ −) 24), “0111b” (−55 ≦ D ≦ −45), “1001b” (−71 ≦ D ≦ −56), “1011b” (−87 ≦ D ≦ −72), “1101b” (−103 ≦ D) ≦ −88), “1111b” (−119 ≦ D ≦ −104). Let the third nibble be "| D | -B". However, B is set as follows according to D. “8” (8 ≦ D ≦ 23), “24” (24 ≦ D ≦ 39), “40” (40 ≦ D ≦ 55), “56” (56 ≦ D ≦ 71), “72” (72 ≦ D ≦ 87), “88” (88 ≦ D ≦ 103), “104” (104 ≦ D ≦ 119), “8” (−23 ≦ D ≦ −8), “24” (−39 ≦ D ≦ −) 24), “45” (−55 ≦ D ≦ −45), “56” (−71 ≦ D ≦ −56), “72” (−87 ≦ D ≦ −72), “88” (−103 ≦ D) ≦ −88), “104” (−119 ≦ D ≦ −104).

  Next, 120 ≦ | D |. The first byte is “80h”. Let P be the second byte.

  When N (5 to 255) of the same D continues, the first byte “18h” and the second byte are N following each encoded value.

  As shown in 4 of FIG. 1, after encoding all of R, these encodings are a mixture of nibble values (4 bits) and byte values (8 bits), so in order from the beginning of encoding. It becomes byte value. Therefore, the data encoded as bytes is also separated into an even-numbered nibble value and the next odd-numbered nibble value, and there is a part that is byteized.

  As shown in 5 of FIG. 1, G and B are encoded in the same manner, but encoding is performed only when the bit table is different from R when viewed from the bit table. Each RGB data is composed of variable length data. First data, encoded data of R difference value. Second data, G bit table. Third data, encoded data of G difference value. Fourth data, B bit table. Fifth data, encoded data of B difference value.

Claims (2)

In a medical color image compression method for compressing and encoding a digital color image taken for medical use,
A first step of separating each pixel data into RGB;
A second step of calculating difference data of adjacent pixels for each of RGB;
A medical color image compression method comprising: a third step of selecting an arbitrary element from RGB and comparing and encoding the difference component and the difference component of the remaining two elements .   In the second step, the difference data is all represented by 4 bits, and the numerical value is composed of 1 bit of code and 3 bits of data, or is composed of a marker for distributing information to the next 4 bits. The medical color image compression method according to claim 1, wherein:

Images