The Study On Robust Feature Extraction Of Infrared Face Recognition

The requirement for reliable person identification in computerized access control has resulted in an increased interest in biometrics. Face recognition, being passive and non-intrusive, has become a very active topic in computer vision research. Even though face recognition can get good performance in well aligned, well illuminated, and frontal pose situation, building an automated system that accomplishes such objectives is very challenging. The challenges mainly come from the large variations in the visual stimulus due to illumination conditions, viewing directions or poses, facial expression and disguises. To tackle with those challenges, infrared images have been used for face recognition. Infrared face recognition, being independent of external illumination and disguises, has become an important direction of computer vision research. However, infrared face recognition has its disadvantages:low resolution and instability due to ambient environment.In this thesis, based on analysis the characters of infrared face image, we get the main challenges of infrared face recognition:1. The low resolution of infrared face image. Due to the limitations of infrared camera, the infrared face can not reach the same high resolution as that of visible face.2. The instability of infrared face image. It is highlighted that the infrared images which are the character of the human skins can be affected by ambient temperature, psychological, as well as physiological conditions.To tackle with those challenges of infrared face recognition, this thesis do some researches in following two aspects:1. Based on the traditional face feature extraction methods and new characters of infrared image, a series of robust infrared face features extraction algorithms are studied.2. Based on the principle of bio-heat transfer, how to transform the infrared temperatures to stable biological information is exploring.The concrete contributions are listed as follows:1. Due to low solution of infrared face images, to extract more region features, the improved holistic feature extraction methods are studied. The main novelty is that the blocking and weighted idea is integrated to the traditional holistic feature extraction methods.2. A novel infrared face feature extraction method based on local derivate binary pattern is proposed. This method has two improvements:extracting two-order direction structure information and distinctive pattern choosing.3. A new construction of blood perfusion is proposed based on bio-heat transfer is proposed. According to this method, we get the stable blood perfusion for recognition.4. A robust infrared face system based on fusion of block-holistic features and local representation features is developed.