The study on low-quality images Geometric Facial Feature Extraction

Abstract

Face recognition method is a heatedly discussed topic in the field of current pattern recognition and artificial intelligence. Although the face recognition technology has made great achievements, most of the face recognition technologies only focus on the high-resolution images in algorithm design and model training, because the human face images have different sources, some images exhibit very poor resolution, such as fuzzy, high noise and low resolution, which have increased the difficulty of image recognition. Based on the characteristics of low-quality images, this paper first uses wavelet transform to preprocess the images to improve the extraction accuracy of geometric features, adopts the geometric feature normalization method, compares the different selection methods’ pros and cons for the face image geometric feature, obtains the corresponding selection method, and effectively improves the recognition effect. This paper describes the experimental analysis of algorithm in the C++ platform. The results show that the algorithm solves the failed recognition of low-quality face images, and improves the face recognition capabilities towards such images. Keywords—Fac Recognition; Low-quality Image; Geometric Feature; Wavelet Transform; Feature Extraction