The Research And Design On Robust Algorithm To Extract The Multiple Face Feature Based On Multiwavelet

Due to the natural and high acceptability, the face recognition technology iswidely used in fields such as video surveillance, judicial application and identification.In recent years, the face recognition technology has developed a lot along the deeperresearch on. However, illumination variation is one of the key factors which haverestricted the popularity of the face recognition technology.Now, there are many algorithms have been proposed to process the effects of thelight. Illumination invariant extraction method with higher performance and smallercomputational complexity has become one of the mainstream methods, and itsrepresentative is the illumination invariant extraction method based on wavelettransform.After further research on the illumination invariant extraction algorithm, wavelettransform theory, dual tree complex wavelet transform theory, the illumination effecton face recognition and histogram equalization, the thesis proposed an invariantextraction algorithm based on dual tree complex wavelet transform andBayesian-denoising model. And with the use of the histogram equalization, this thesisimproved the illumination invariant extraction algorithm based on wavelet transformand Bayesian-denoising model(WBD algorithm) and the illumination invariantextraction algorithm (DBD algorithm) based on dual tree complex wavelet transformand Bayesian-denoising model, and get the improved illumination invariant extractionalgorithm (WBDH algorithm) based on wavelet transform and Bayesian-denoisingmodel and the improved illumination invariant extraction algorithm (DBDH algorithm)based on dual tree complex wavelet transform and Bayesian-denoising model.The main work and innovation of this thesis are as follows:1. Proposed the WBDH algorithm. Histogram equalization will enhance thecontrast and hue of the image. The illumination invariant mainly contains the reflectioncoefficient of the image, and it contains less light components, so the contrast of theimage is low, the edge info is also not obvious enough. The thesis enhanced theillumination invariance extracted by WBD algorithm with the histogram equalization, and achieved good effects. The experimental data shows that the illumination invariantenhanced by histogram equalization can effectively improve the correct rate of facerecognition algorithm under the complex light conditions.2. Proposed the DBD algorithm. The dual tree complex wavelet transform is a newwavelet technology. It is the development to the traditional wavelet technology, andhas many excellent characteristics which are beneficial to the image processing. Basedon this, we propose to use the dual tree complex wavelet transform to improved theWBD algorithm and get the DBD algorithm. The performance of the algorithm has beenimproved on Yale B face database.3. Proposed the DBDH algorithm. This algorithm is also an improved algorithmbased on DBD algorithm with the histogram equalization. The experiments show thatthe DBDH method has the best recognition rate in all algorithms.