Algorithm for extracting texture features of curvelet transformation based on energy distribution

Abstract

The invention discloses an algorithm for extracting the texture features of curvelet transformation based on energy distribution and relates to the technical field of digital image processing. A dual-layer curvelet transformation module separately calculates mean values and variances of 13 sub-bands, arranges 52D original feature vectors in such an order: a first layer of high frequency sub-band, a second layer of high frequency sub-band, a low frequency similar sub-band. A main direction extraction module calculates proportion of each sub-band in the total energy of the layer where each sub-band is arranged in first layer and second layer high frequency parts, arranges high frequency energy distribution in order, finds the sub-band having the highest energy part of each layer, and refers the sub-band that has the highest energy part as a reference sub-band. A circulation translation module conducts circulation and translation on the entire original feature vector until the reference sub-band is moved to a head part of a feature vector, and after the moving, the front two components of the feature vector of each image are the mean values and variances of the reference sub-band. According to the invention, the algorithm can better analyze complex textures, overcomes the defects of absence of rotation invariance of traditional curvelet transformation algorithm, increases precision, and is conductive to image searching.

Description

technical field [0001] The invention relates to the technical field of digital image processing, in particular to an energy distribution-based curvelet transform texture feature extraction algorithm. Background technique [0002] The texture feature reflects the change of the gray value of the image. It is one of the important underlying features of the image. It is the own characteristic of the object material itself and does not change with the outside world. After summarizing by many researchers, it is concluded that texture features are mainly divided into four categories: statistical method, structural method, spectral method and model method. In spectral method, wavelet transform algorithm and curvelet transform algorithm are represented. [0003] The wavelet transform algorithm, which is a commonly used multi-scale image analysis method, can extract texture features from images, but the wavelet transform coefficients have many defects, such as no rotation invariance, ...

AI Technical Summary

Problems solved by technology

[0003] The wavelet transform algorithm, which is a commonly used multi-scale image analysis method, can extract texture features from images, but the wavelet transform coefficients have many defects, such as no rotation invariance, sensitivity to translation, and few analytical directions. When the input query When the example image is rotated or translated, the retrieval accuracy is significantly reduced

[0004] The curvelet transform algorithm, the curvelet transform derived from the wavelet transform has very good characteristics in both the space domain and the frequency domain. The curvelet transform has good resolution for curves and complex textures. More than eight, it is more suitable for retrieving the current image database than the wavelet transform, but the curvelet transform still cannot overcome the drop in precision caused by image rotation, the algorithm aims to realize the image texture feature extraction method with rotation invariance, based on this , it is necessary to design a new energy distribution based curvelet transform texture feature extraction algorithm

Images