IMAGE FEATURE EXTRACTION BY MULTIRESOLUTION ANALYSIS WITH WAVELET TRANSFORM
This thesis is a study of the 2D-image signal feature extraction by Multi-Resolution Analysis (MRA) with Wavelet Transforms (WT). The Wavelet. Transform is the most significant signal analysis tool developed in last decade and it is extensively used in several fields such as signal processing, image processing, and communication. This thesis is devoted to the development of a novel feature extraction scheme for image analysis with WT. The goal is to utilize the merits of WT in feature extraction for stereo image matching, given that the image features involved are not well pre-defined. The concept of Multi-Resolution Analysis, which is inherited naturally in WT, is employed in the decomposition of an image. The image properties at each resolution level are represented by the scalars of a wavelet and its translations at that level; in tum the feature space is formed by the wavelet at each decomposition level, precisely, by the wavelet and its dilations. A novel WT algorithm, the Anchored Wavelet Transform (AWT), for MRA is introduced, and the usefulness of the new algorithm for feature extraction is demonstrated with examples. With the feature extraction scheme proposed, a stereo image matching algorithm is coded which could be a general pre-process step in 3D-object parameter estimation. Other usage could be found easily in image or texture pattern classification, surface segmentation and object tracking in an image sequence.