Face recognition feature extraction algorithm based on totally symmetric local Weber descriptor

Abstract

The invention relates to a face recognition feature extraction algorithm based on a totally symmetric local Weber descriptor. The algorithm has the main technical characteristics that a gray difference value between a neighbourhood pixel and a target pixel in a 3*3 local window is calculated to reflect face image local gray difference information, an absolute value sign is added to describe a local gray change degree, and the texture information change of the image in different gradient directions can be embodied through considering the spatial position relationship of a surrounding pixel in avertical direction, a horizontal direction and a diagonal. The difference of the gray values between the surrounding pixel and the target pixel is considered, and the spatial distribution direction information characteristic between the surrounding pixels is also considered. The algorithm is applied to a thermal infrared face database, more effective texture detail characteristics with better discriminability can be extracted, a face identification rate is obviously improved, and the algorithm has good robustness and generalization ability and can be widely applied to image processing fields,including thermal infrared face recognition and the like.

Description

technical field [0001] The invention belongs to the technical field of image processing and biological recognition, in particular to a face recognition feature extraction algorithm (CSWLD) based on a fully symmetrical local Weber descriptor. Background technique [0002] As a core technology of face recognition, face feature extraction has a vital impact on the performance of face recognition systems. In the past 20 years, many face feature extraction methods have been proposed, which can be mainly divided into two methods based on global features and based on local features. [0003] The method based on global features mainly counts the overall feature attributes such as the variance, color, and histogram of the image. Representative methods include principal component analysis (PCA), linear discriminant analysis (LDA), etc. PCA does not consider the sample category, so the resulting low-dimensional space is not optimal for discriminative classification. The LDA method ca...

AI Technical Summary

Problems solved by technology

In 1994, T.Ojala et al. of Oulu University in Finland proposed the Local Binary Pattern (LBP) to describe image texture features. This method is invariant to illumination changes, but the types of binary patterns will increase sharply as the number of sampling points increases. The dimension of the feature space is too high or even "dimension disaster", and it is sensitive to noise, and the problem of posture and expression is not solved.

Lades et al. first used Gabor wavelet to represent face images. Gabor wavelet is sensitive to image edges and can provide good direction selection and scale selection characteristics. However, it has the problem of weak ability to represent global features and redundant feature data.

The Weber Local Descriptor (WLD) feature proposed by Chen et al. can effectively represent image texture features, is more resistant to noise interference, and weakens the influence of illumination changes. However, the WLD method only uses the contrast information between the surrounding pixels and the central pixel in the field. Does not take advantage of the information relationship between surrounding pixels

Later, some scholars improved the WLD algorithm from different angles, but these algorithms only calculate the gradient information in the horizontal and vertical directions, and the utilization rate of the spatial structure distribution information is low.

[0005] To sum up, the existing local feature extraction algorithms are still difficult to accurately describe the facial features, and the recognition rate cannot be effectively improved.

Images