Quantitative Analysis on LGSD and HE Based Image Block Division Schemes
Tian‐Yun Huang · 2018
Image block division or grain partitioning plays an important role in many image applications, such as visual coding, region segmentation, rough set representation, grain computing, fuzzy decision, etc. Since the conventional fixed-size block scheme cannot capture the topology structure and grayscale distribution in the image, it always performs not well. A hyper entropy (HE) based image block division scheme is proposed in this paper, and it is compared with the local grayscale standard deviation (LGSD) based and fixed-size block division schemes, from perspectives of algorithms, optimality and performance. Experiments on LGSD have shown that the search is always degraded to some local optimal solution on non-convex surface, thus it performs only relatively better than the fixed-size scheme. On the contrary, HE based scheme captures the grayscale distribution rather well in most images, since the global optimal solution can always be reached. The full quantitative analysis and quality evaluation on some typical images validate its effectiveness.