Image feature extraction algorithm based on bi-dimensional local mean decomposition

The scale invariant feature transform (SIFT) algorithm has been applied to many fields, it has been found that the algorithm has some problems, such as high complexity, it is easily led to the dimensional disaster and it is not completely affine invariant. Some scholars have proposed improved algorithms to solve these problems, but these algorithms were specific to certain solutions and not applicable for comprehensively solving the above problems. To solve these problems, an adaptive image feature extraction algorithm based on bi-dimensional local mean decomposition (BLMD) and SIFT is proposed in this paper. First, adaptive BLMD is used to decompose the image and obtain a number of bi-dimensional production functions (BPFs). Second, the SIFT algorithm based on parameter optimization is used to extract the features of the decomposed BPFs. Finally, we synthesize and process the feature information extracted by the BPFs to obtain all the feature information of the original image. Traditional feature extraction methods and the proposed method are compared and analyzed in three different scenarios involving face database images with different scales and levels of blur. The proposed method yields rich and complete feature information and is beneficial to image matching and registration. Moreover, the proposed is more efficient than other methods. The proposed approach provides a technical method for image adaptive feature extraction and a directional framework for the development and improvement of adaptive image feature extraction schemes.