Face Tracking Research Under Low Illuminance Condition

Face tracking is an important research topic in computer vision and pattern recognition. It has great value in many areas, such as: content-based video retrieval, digital video processing, human computer interface, video surveillance and so on. But in the practical application, face tracking system usually encounter the problem of different illumination circumstance or insufficiency of light. Based on the thorough knowledge of the method in face tracking and image enhancement techniques, this thesis proposes some new methods. The main contents of the thesis are as follows:1. A novel Retinex algorithm based on bilateral filtering is proposed to the question of the apparition of halo artifacts in the presence of high contrast edges using exiting Retinex image enhancement. Firstly, this algorithm decompose the original image into the illuminant image and the reflectance image using the bilateral filter which has the function of edge-preserving, then using different strategies to compress the illuminant image and to enhance the reflectance image. At last, it combines the two part images to a new image. This algorithm is verified from subjective visual effects and objective image quality assessment, experimental results show that it has good enhancement result and can avoid halo artifacts.2. Based on the human vision property, improved from the center/surround Retinex, A novel algorithm is proposed to the question of the noise magnify of the original image. Firstly, this algorithm use orthogonal Prewitt operator to calculate the grads image, then convolve with the 5×5 Gaussian kernel function and transform the result with a linear function. At last, use this final result as the spatial activity matrix of the original image multiply the surround matrix of the Center/surround Retinex algorithm. Experimental results show that it is effective and has a good enhancement result.3. Practical face tracking system usually encounters the situation of different illumination circumstance, which will cause the color distortion of the human face. It will limits the usage of skin color based method in face tracking system. By introducing our proposed method mentioned above to the real system, experimental results show that it can effective eliminate the limitation.