Face Recognition Based On Wavelet Neural Network

Wavelet Neural Network(WNN)which is combined wavelet analysis with Artificial Neural Network has been widely researched in pattern recognition field.This article - launches the research to Sichuan Education Department Foundation Applied Research Project Based on the Methods Research of Synthesis Intelligence Pattern Recognition (2005A140). A detailed discussion on WNN theory has been made, and WNN has been studied thoroughly in the project. At the same time, WNN is applied in face recognition. A face recognition method, which is based on Independent Component Analysis (ICA) and WNN, is proposed. The WNN construction methods, the WNN training methods as well as the WNN structure optimization methods are especially studied.The facial images are preprocessed before better ICA base and facial feature are extracted. Then WNN that can avoid the blindness of the design is constructed under the direction of the wavelet theory. Afterward, we train WNN by facial features which is abstracted by the better ICA base. When testing, we input facial features, which is abstracted by the better ICA base to the WNN that has been trained to accomplish face recognition.This article occupying of innovation to:(1) We use the smallest distinction entropy theory to choose appropriate biorthogonal wavelet base.(2) We use the structure determination and the parameter estimate method of NARMAX model to optimize network architecture and to initialize weight coefficient.(3) We improve rate adjustment algorithm in WNN training process. Otherwise, we propose a more reasonable rate adjustment method to make it better easy to convergence. Experiments based on the ORL face database indicate that algorithms proposed here could have a better representation of the differences among faces. Simultaneously, the WNN, which is constructed by the algorithms proposed here, can greatly enhance the anti- noise performance of the WNN as well as greatly reduce operative complexity. The algorithms proposed here can enhanced the face recognition rate effectively too.