A Detailed Explanation of PNG Compression and Decoding Encoding

The PNG compression process consists of two stages: Prediction and Compression. In the Prediction stage, the image data is processed line by line. Each pixel's channel values (such as ARGB) in a row are first processed by the Filter stage, then recalculated by a differential processor through differential encoding. This encoding calculates the difference between a pixel's channel value and the corresponding channel value of the previous pixel (either horizontally or vertically). When adjacent pixels have similar channel values, the result contains many small values like 0, 1 or -1.

The key objective of the Prediction stage is to select the optimal differential processor to maximize the number of zeros and repeated values in the encoded output, which directly affects the compression ratio in the next stage. It's important to note that the differential encoder compares individual channel values between pixels, not entire pixels.

After Prediction, the transformed data is passed to Deflate for actual compression using LZ77 and Huffman coding, with the final compressed result being stored. The compression efficiency depends on two factors: First, the Prediction results - regions with similar colors (producing many zeros) achieve higher compression, while areas with significant color variations show poorer compression. This means for high-resolution images where lossless compression isn't essential, JPG format may be preferable. Conversely, for simpler images requiring transparency support, PNG is superior to JPG. Second, Deflate's row matching performance - since processing is done row by row, Deflate limits matching symbol counts to between 3 and 258, with a maximum compression ratio of 1032:1. When fewer than 3 symbols match, compression may fail, meaning changes to image width can affect compression efficiency.