Topics in Statistical Image Analysis

Abstract

There are plenty of interesting problems in image processing that can be solved by statistical methods. In this talk, I will introduce the following examples from my recent research work. (i) OCR (Optical Character Recognition) is the process of translating images of texts into computer-editable texts. I will present a brand new algorithm for this translation. The algorithm first uses a technique for image partitions. Then it uses Kullback-Leibler distance for image classifications. (ii) Motion Estimation is the process of analyzing a series of image frames to identify motions of objects. The block-matching method with L1 norm is commonly used for this sort of problems. I will introduce a coarse-to-fine scheme that can dramatically reduce the computations, while also guarantees the correct solution. (iii) Image DeWarping targets to remove distortions in images. This problem can be transformed to the problem of seeking an optimal mapping from the actual observed frame into the desired ideal frame. The Thin Plate Spline is an effective tool for providing such a solution. However, it may introduce artifacts. In this example, I will present a method that modifies the Thin Plate Spline to reduce artifacts.