壓縮感知人臉識(shí)別`

1、中國(guó)科學(xué)技術(shù)大學(xué)碩士學(xué)位論文壓縮感知人臉識(shí)別`姓名：平強(qiáng)申請(qǐng)學(xué)位級(jí)別：碩士專(zhuān)業(yè)：信號(hào)與信息處理指導(dǎo)教師：俞能海；莊連生2011-05-02ABSTRACT III ABSTRACT As one of the hottest research directions in the field of computer vision and bioinformatics, face recognition has provided power

2、ful technical support for the informatization process the core applications in public security and human-machine interaction, which has been capturing special attention from both academia and industry. Having devoted muc

3、h effort and spent a large amount of funding on face recognition, the researchers are still not satisfied with its slow development in the past decades. Even today, further development of face recognition is still seriou

4、sly challenged by complexity of illumination, variability of pose and expressions, and randomness of occlusion. Most of the existing face recognition algorithms are based on the classical statistical learning theory,

5、which has been proved to be effective in solving low dimensional problems where sufficient training samples are available. However, the classical statistical learning theory cannot well handle problems with high dimensio

6、nal data due to the characteristic of face images. Meanwhile, the number of collected training samples is severely restricted in real applications. Due to the above aspects, the classical statistical learning theory is n

7、ot very suitable for face recognition applications. In 2006, Donoho and Candes proposed a novel framework named compressed sensing (CS). This framework has aroused another upsurge in face recognition since it’s introduce

8、d into the face recognition area, and one of the most outstanding algorithms based on CS is Sparse Representation-based Classification (SRC). Compared with most existing algorithms, SRC implements statistical inference b

9、y directly exploiting the sparse distribution of high dimensional data, which can handle the curse of dimensionality effectively. Moreover, SRC implements face recognition via image pixel values and avoids information lo

10、ss thanks to the pre-processing procedures. SRC requires exact alignment between each test image and training images. Nevertheless, variation of poses and expressions leads to the error on alignment and thus the SRC’s pe

11、rformance may decrease. This fact severely restricts SRC’s generalization ability to real-world face recognition problems. This dissertation focuses on the research of CS-based face recognition algorithms that handle pos