Statistical color texture descriptors for histological images analysis

Abstract

In this paper we compare different approaches to combine color and statistical texture descriptors. Previous studies on this topic were conducted on natural images only. We focus on the particular case of histological datasets where color plays an important role due to the staining process of the biological samples. We also introduce two new variants of the well-known Local Binary Patterns (LBP) operator. We test these approaches on three diversified histological datasets. We show that combining color and texture features extracted separately is preferable on datasets having a large variability in the staining, while simultaneous extraction of color and texture information is recommended only for more standardized stainings.