Há alguns posts interessantes aqui que parecem descrever por que A compactação progressiva pode ser melhor e parece que se trata do fato de que, no JPEG progressivo, os dados compactados são ordenados com mais eficiência e que os dados e o ruído da imagem em bloco são separados e compactados separadamente.
Eu destaquei as peças importantes abaixo, elas descrevem melhor do que eu jamais poderia.
Como resultado, eu esperaria que o JPEG progressivo fosse uma compactação melhor , mas nenhuma mudança significativa na qualidade da imagem.
Progressive JPEG has two different coding features:
- Spectral Selection
- Successive Approximation
What you describe is just the first feature (first DC, then AC bands). However, it is the second feature which is the major reason for different coding efficiency of Progressive JPEG. With Successive Approximation, you first store the higher bits of the coefficients (low precision, coarse image), then the lower bits (high precision, detailed image). In usual images, the noise is concentrated in the lower bits. Therefore, the Huffman encoder can be more efficient in symbol coding in the higher bit range where less noise is. The Successive Approximation coding thus separates the noise from the image, and that is the reason for better coding efficiency (noise is hardly compressible).
The more noise (detail) the image has, the better the coding efficiency of Progressive JPEG. The less noise (blur) the image has, the better the coding efficiency of Sequential (Non-progressive) JPEG.
Note that most Progressive JPEG coders use a mixture of Spectral Selection coding and Successive Approximation coding. For better coding efficiency, mainly Successive Approximation is relevant, and you can try to find optimal Successive Approximation paramters (the point where to separate the precision - Ah/Al in the JPEG standard) for your class of images. For example, I sometimes found an optimal point Al=4 for my images (4 lower bits separately coded).