JPS644390B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention relates to image data compression encoding/decoding. The images targeted by the present invention are a plurality of images having a high correlation. For example, first of all, it is a set of multiple images corresponding to the wavelength band obtained from the reflected signal of the subject of electromagnetic waves, including light of different wavelength bands, taken at the same time. An example of this is a multispectral scanner image (hereinafter abbreviated as MSS image) taken from an aircraft or an artificial satellite. The second is multiple tomographic images taken at spatially close positions in three-dimensional space. This can be given as an example. There is a third basic image, and a set of multiple images having the same attributes as that basic image, for example, if a typical lung X-ray image is used as the basic image, any lung X-ray image is an example. An example is a set of multiple images taken in the same area at different times. These images are named strongly correlated images. Now, among these images, we will explain the multispectral scanner image from the Landsat satellite as an example. Landsat satellites have sensors that are sensitive to light in four different wavelength bands.
The four types of images obtained by each sensor are referred to as a band 4 image, a band 5 image, a band 6 image, and a band 7 image in order of the shortest center wavelength. Each band consists of 3240 x 2340 pixels, each pixel has 8
Quantized into bits. In other words, it has a capacity of 7.56 Mbytes per image for one band, and 30 Mbytes for four bands. Dozens of images are transmitted per day, and each ground station stores them as they are on magnetic tape. Furthermore, since it is desirable for receiving stations to store all image data from the past in order to meet the demands of each user, the amount of storage is constantly increasing. Furthermore, in order to prevent memory loss due to changes in the magnetic tape over time, it is necessary to periodically read and write data to and from all magnetic tapes, which is very time-consuming. Next, regarding processing after receiving Landsat image data, it is necessary to transmit the data to a data distribution center after it has been digitally processed and transmitted to another ground processing station capable of digital processing. This amount of transmission is also enormous. Several proposals have been made to efficiently store or transmit these images. As a data compression method for storage, for example, the 1971 National Conference of the Institute of Electronics and Communication Engineers, p.
72, No. 991, ``Compression of LANDSAT image data,'' includes a method of taking the difference from the previous pixel data in the line direction within the band and Huffman encoding the difference value. In other words, the next pixel value is predicted using the data value of the previous pixel in the line direction, but this prediction method is difficult to use for images that contain many high frequencies, that is, images that change rapidly. It has the disadvantage that the rate is poor. As for data compression for transmission, band compression encoding for transmission of TV images and still images is a well-known technique. However, this technology is a data compression encoding method that allows a certain degree of error, and considering that strongly correlated images are the type of images that are not only evaluated by humans but also analyzed by computers, there are no errors. However, using this method is not a good idea as it does not satisfy the user's needs. Huffman coding is well known as a code construction method that is error-free and has a short average code length. However, Huffman encoding is a code that achieves the shortest code length when there is no correlation between sample points, but has the drawback that the above cannot be said when there is correlation between sample points. ing. An object of the present invention is to eliminate the drawbacks of the prior art described above in order to solve the difficulties in storing or transmitting image data described above, and to provide a method for compressing and encoding inter-picture linear predictive data without errors and having high compression efficiency. The aim is to propose the following. Next, the principle of the present invention will be explained. If we define the correlation between images as shown in Equation (1), an example is a Landsat image (an image of the Kanto region taken by Landsat 2 on November 26, 1972).
The inter-band correlations between bands 4 and 5 and between bands 6 and 7 are shown in Table 1, showing high correlation values. Hereinafter, the configuration of the present invention will be shown using this image.

[Table] In addition, xi indicates the i-th pixel value of the reference image, and yi
indicates the i-th pixel value of the predicted image. Using this correlation, band 5 to band 4,
Then, band 6 to band 7 are linearly predicted using equation (2) as an example. yi^=axi+b (2) Here, yi^ is the predicted value of the i-th pixel of the predicted image. a and b are optimal coefficients, which are determined as a function of the correlation value R shown in equation (1) by solving a well-known optimization problem that minimizes the square mean of the true pixel value and the predicted value. The difference between the true pixel value and the predicted value (hereinafter referred to as prediction error) is encoded, but in the case of Landsat images, when we examine the prediction error, we find that if the prediction error shifts in the positive direction, a large number of pixels If there is a shift in the positive direction and a shift in the negative direction,
It can be seen that there is a shift in the negative direction for a considerable number of pixels. This is because the prediction coefficients are determined for a fixed area and cannot follow the partial characteristics of the image. Note that a prediction method that follows the partial properties of an image can be considered, but the processing would be extremely complicated and would be impractical. The change probability of positive and negative signs is as shown in Table 2 when actually measured for Landsat images. Note that the positive and negative signs representing plus and minus are expressed as positive and negative signs.

[Table] Therefore, in the present invention, we focus on changes in the positive and negative signs of the prediction error, and for pixels where the positive and negative signs of the prediction error do not change, Huffman encoding is performed using the absolute value of the prediction error as an example. For pixels whose sign changes, encoding is performed by adding a positive/negative sign change sign to the sign of the absolute value of the prediction error. Table 3 shows an example of the configuration of the code of the present invention.
5 shows a code sequence according to the code structure in Table 3 for an appropriate prediction error sequence. For comparison, Table 4 shows the simple Huffman code, and Table 6 shows the simple Huffman code.
The code sequence according to Table 4 for the same prediction error sequence as in Table 5 is shown.

【table】

Claims (1)

[Claims] 1. When storing and transmitting multiple images with high correlation between images, the pixels of the image to be encoded are predicted by a linear combination of the pixel values of the corresponding pixels of the reference image and its neighboring pixels, and the prediction error is encoded. In data compression encoding, for pixels where the sign of the prediction error does not change, the absolute value of the prediction error is encoded, and for pixels where the sign of the prediction error changes, the absolute value of the prediction error is encoded. An inter-picture linear prediction data compression encoding method characterized by performing encoding by adding a positive/negative sign change code to the code.