| Vision is a challenging research subject in the field of science and engineeringand is of great theoretical significance and broad applications. General ObjectRecognition is one of basic problems in computer vision and is hard to deal with.Although many algorithms and systems for specific purpose like face detectionand recognition have been developed and widely used, developing a general sys-tem for general object recognition is far from mature.Aiming at this problem, a novel feature composition model for general ob-ject recognition based on probabilistic method and scale space theory has beendeveloped in this paper. Firstly, a survey on recent achievements of general ob-ject recognition has been made, and detection and recognition algorithms basedon patches have been analyzed in this paper. A novel hierarchical feature compo-sition model is then proposed , and the implementation algorithm is developed.At last, computer simulations are given to evaluate the performance of the algo-rithm, and some comparisons are provided.The main contributions of this thesis are as follows:1. From the view of biological theory and cognitive process, the necessity offeature composition is analyzed and existing feature composition algorithmsare introduced. After that, a novel concept of feature composition basedon scale space theory is proposed.2. Based on the feature composition concept, a novel combination methodnamed"stable triplet"is introduced, which combines small patches to present high-level features. Using the stable triplet, hierarchical featurecomposition model is designed to solve general object recognition problemwithout using extra knowledge for specific object.3. Practical modeling and computer implementation for the hierarchical fea-ture composition are provided in this paper. Approximation algorithm isdeveloped in order to reduce the computational complexity. Additionally,simulation experiments are made to evaluate the performance of the pro-posed method.. |
PDF Full Text Request