scientific article; zbMATH DE number 4041776

From MaRDI portal
Publication:3780982





[For the entire collection see Zbl 0632.00021.] Performing a lossless encoding of gray scale images by applying prediction methods with additional Huffman coding of the differences results in long codewords for a small fraction of pixels, especially those representing the contours of the image. The overall coding method might be improved by joining such pixels with the 3-bit chain code and by specifying them by pulse code modulation instead of their Huffman codeword. The problem of maximizing the possible gain of such a technique leads to the theoretical question of maximizing the expression \(\sum^{n}_{i=1}p_ i\cdot (l_ i-g)_+,\) where \(a_+:=a\) for \(a\geq 0\), and \(a_+:=0\) for \(a<0\), with a specified number n of codewords and a given value \(g\in {\mathbb{N}}\) over all probability distributions \((p_ 1,...,p_ n)\) and their codeword lengths \((l_ 1,...,l_ n)\) according to the Huffman procedure. For this question an approximate solution which is nearly optimal is found. The application of this solution to the encoding of digital images with 128 grey scale levels makes sure that the possible gain of the described method is bounded by 0.2005 bit/pixel. It is quite natural that the results of the performed test with four real images are significantly lower than the theoretical maximum.












This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3780982)