Basic principles of digital image compression

The existence of redundant information is the basis of image compression, and the process of compression coding is to remove all kinds of redundant information of the image, which mainly includes the following parts:

• Spatial redundancy: Adjacent pixels in images and videos have similar or identical pixel values.
• Temporal redundancy: There is a large correlation between adjacent frames in the video image.
• Coding redundancy: It indicates that the average number of bits of a digital image pixel is greater than the information entropy of this digital image.
• Structural redundancy: Some images have strong texture structure in some regions or strong correlation between regions. Knowing one part of the information in the image can infer another part of the information.
• Visual redundancy: Due to the biological characteristics of visual masking, the human eye cannot perceive or is not sensitive to some information in the image, which constitutes visual redundancy.
• Inter-spectral redundancy: There is a strong correlation between pixels of adjacent spectral segments in hyperspectral images.
• Quantization redundancy: Using fewer quantization levels to quantize the original image signal without affecting the subjective visual quality.

In principle, the compression process has three basic links: transform, quantization and encoding.

Transform

Transformation mainly means that the correlation between image pixels is not only in the static position relationship, but also in the frequency domain, so the problems that cannot be solved in the spatial domain can be solved in the frequency domain.

Quantization

Quantization is mainly applied to the conversion from continuous analog signal to digital signal, and its role is to approximate the continuous value of the signal into finitely many discrete values. Quantization is mainly to quantize the transform coefficients after the transformation, that is, to express more transform coefficients with less quantization values, so as to further compress the data.

Encoding

The goal of entropy coding is to remove the information redundancy of source symbols, which is also called information entropy redundancy or coding redundancy. Entropy coding is generally at the end of the system, which is responsible for the entropy coding of transform coefficients, motion vectors and other information generated in the coding process, and the organization of the final encoded bitstream.

Your task

What is the basic process of image compression?

Share your thoughts and ideas in the comments below.