For Target Recognition, Image Feature Fusion Technique Of Extracting

Automatically recognizing objects from images is in a lot of demands and has great prospect in civilian applications and national defense, hence it is an important area of information processing. Since there are difficulties to obtain accurate and reliable results from single image and a feature, it is a necessary way to adopt information fusion technology, which has been a hot point research area.One important problem in image-based object recognition is to extract image features exactly and reliably. The objects in an image take some forms of the sets of pixels with certain attributes. The features describing the objects are the sets of pixels and their attributes, which are named symbols and attributes. They form the basis of object recognition. This dissertation focuses on the concepts and methods for feature extraction for object recognition by means of information fusion technology, and involves following three respects.The symbol extraction methods by fusing different image information are studied, and the concepts about symbol extraction and its issues are summarized. Two region segmentation methods by fusing multiple features are presented according to different viewpoints. The first method fuses multiple attributes of a region in order to expand the measures describing the consistency of regions. The second method fuses two different kinds of symbols, which are lines and regions. The experimental results indicate that using the complementarity among the different symbols and the attributes improved segmentation performance.The concepts and methods about symbol extraction from multi-sensor images are studied. For the fusion of lines, one complete mathematical concept as well as some concrete fusion operators definition are presented. Two mathematical definitions of the extraction of parallel lines from multi-sensor images are presented. And a method of extracting road network based on multi-band remote sensing images is presented. The experiment verifies their effectiveness while the methods are applied to actual images.Taking range-intensity images as example, the extraction of symbols by fusion of images of different sensing mechanism is studied. After summarizing the existed methods about fusing the two images, a region segmentation method based on physical imaging model and multiple image features is presented. And the experimental results are satisfying.