Joint Photographics Expert Group (JPEG) is both a standardization body and data compression method and a graphic file format. Since 1987, a working group of ISO/ IEC SC29/WG10 has been working on the standardization of a method for the compression of color images and digital photos. The process for compressing pixel graphics is a lossy compression in which information loss is deliberately accepted in order to achieve a higher compression rate.
The aim of the JPEG activities was to develop a compression algorithm that was geared to application-specific requirements. This includes, for example, the possibility for the user to display reproduced photos in different quality levels. In addition, JPEG should be suitable for images with color gradients and have no limitations in terms of image size, color depth, pixel format or complexity and color variety. In addition, one requirement related to the manageability of the process so that it could be used in many different hardware platforms.
JPEG compression proceeds in several stages: First, the primary colorsRGB of the pixel graphic to be compressed are decomposed into their brightness component, luminance (Y), and their color component, chrominance (C), and converted to the YCbCr color model. Using color subsampling, the YUV or YCbCr signal is subjected to an initial data reduction. The color image is then divided into pixel blocks for brightness and color signals. This is referred to as downsampling because two, four or eight neighboring pixels are averaged in color value and combined into one pixel.
In the following Discrete Cosine Transformation( DCT), the intensity data is transformed into frequency data. In this process, the individual 8x8 pixel blocks are translated channel by channel into their frequency components and a coefficient matrix is created from this.
This matrix contains 64 DCT coefficients representing the proportion of each function in this representation. It contains as much or even more data as before the DCT transformation, because the individual spectral values are represented with higher precision due to the avoidance of rounding errors. However, this matrix data can be reduced without visibly affecting the image impression. Since the human eye no longer recognizes small changes in brightness in fine image details, the high- resolution image details are quantized with lower resolution.
Since the DCT transform itself does not cause data compression, the coefficient matrix is quantized. In this quantization, each coefficient is divided by a certain factor and then rounded, smaller values are eliminated. The divisors for the coefficients are stored in quantization tables in the headers of the JPEG file, separately for color and brightness signals. With higher image compression, the quantization values become larger and more DCT coefficients are eliminated. Therefore, the quantized coefficient matrix contains much fewer and smaller values than the non-quantized one.
After quantization, the actual compression takes place, which includes only the difference values of the color components in relation to a base value, since successive values differ only slightly from each other. Since the deviations from the determined reference value are relatively small, JPEG requires relatively little data to compress an 8x8 matrix. These are encoded using Run Length Encoding( RLE) followed by Huffman encoding. The performance of the process is underlined by the quality classes for JPEG encoding.