Adaptive multi-level 2D Karhunen-Loeve-based transform for still images
Roumen K. Kountchev, Kazumi Nakamatsu · International Journal of Reasoning-based Intelligent Systems · 2014
In this work is presented one new approach for block processing of halftone images, based on the adaptive multilevel Karhunen-Loeve (KL) transform. For this, the rows and the columns of the digital image blocks are processed sequentially, using KL matrices of size 2 × 2. As a result, each row of the processed block obtained one vector. The vector components are rearranged in correspondence to their mutual correlation, starting from the highest. After that, on all vectors is applied the next transform level, etc. When the transform for the rows is finished, the processing is executed in a similar way for the columns. The result obtained strong spatial decorrelation of the image blocks elements. The basic advantages of the new algorithm to the famous 2D KL transform are the lower computational complexity and the simplified structure, which offer better opportunities for parallel and recursive image processing.