Research Of Feature Extraction Based On Local Binary Patterns

Texture feature extraction refers to extract the texture feature by image processing and then get the quantitative and qualitative description of texture. Local Binary Pattern is a newly-developed theory which is simple but useful calculating method of texture analysis. In the computer vision field, LBP performs well and is widely recognized.In this thesis, texture feature extraction is applied to recognition of human face and handwritten digits, which improve the problem that there is no sole answer in recognition of human face by the universal k-Nearest Neighbor approach based on LBP. Meanwhile, LBP is also applied to recognition of handwritten digits in this thesis which is rarely noticed now. The consequences approve that LBP is feasible in these fields mentioned above.Main points and innovative proposals are as follows:1. Researches on extracting and categorizing phases of recognition of human faces based on LBP. In the extracting phase, single-blocked multi-blocked and sliding windows are applied to extract LBP features. The recognition rate and efficiency of these approaches are also compared. In categorizing phase, nearest priority is applied to select best answers based on KNN calculating method. Based on Yale face database A, relevant methods and improvements are experimented. By analyzing data in experiments, the methods are approved to be correct and the improvements are efficient.2. Using LBP in the recognition of handwritten digits. The recognition of handwritten digits based on LBP features in MNIST handwritten digits database is experimented and the consequences prove the LBP method is effective.