Local Displacement Estimation of Image Patches and Textons

Abstract

This chapter presents a novel dissimilarity measure for images, called Local Patch Dissimilarity (LPD). This new dissimilarity measure is inspired from the rank distance measure for strings. Hence, it shows the concept of treating image and text in a similar way, in practice. An algorithm to compute LPD and theoretical properties of this dissimilarity are also given. The chapter describes several ways of improving LPD in terms of efficiency, such as using a hash table to store precomputed patch distances or skipping the comparison of overlapping patches. Another way to avoid the problem of the higher computational time on large sets of images is to turn to local learning methods. Several experiments are conducted on two data sets using both standard machine learning methods and local learning methods. The obtained results come to support the fact that LPD is a very good dissimilarity measure for images with applications in handwritten digit recognition and image classification. A variant of LPD, called Local Texton Dissimilarity (LTD), is also presented in this chapter. Local Texton Dissimilarity aims at classifying texture images. It is based on textons, which are represented as a set of features extracted from image patches. Textons provide a lighter representation of patches, allowing for a faster computational time and a better accuracy when used for texture analysis. The performance level of the machine learning methods based on LTD is comparable to the state-of-the-art methods for texture classification.