Texture feature extraction method of local ternary pattern based on mean value sampling

Abstract

The invention discloses a local ternary pattern texture feature extraction method based on mean value sampling. Firstly, the rotation invariance is used to perform preliminary dimension reduction on local ternary coded sequences of all possible values, and then further dimensionality reduction is realized by using dimensionality reduction conditions. Establish a mode mapping table that contains the correspondence between the original mode number and the final mode number; for the texture image to be extracted, perform neighborhood circular symmetric mean sampling on each non-edge pixel of the texture image, and obtain effective coding points from the sampling points Perform encoding to obtain the local three-value encoding sequence of non-edge pixels and the corresponding original mode number, then look up the corresponding final mode in the mode mapping table, and count the number of non-edge pixels covered by each final mode in the texture image , to construct the feature vector of the texture image. The invention controls noise through mean value sampling, improves the accuracy of texture features, and effectively controls the dimension of feature vectors through a dimensionality reduction method.

Description

technical field [0001] The invention belongs to the technical field of texture feature extraction, and more specifically relates to a local ternary pattern texture feature extraction method based on mean value sampling. Background technique [0002] The extraction of visual features is an important link in the process of image classification and recognition, and the quality of features directly affects the performance of the entire visual system. In long-term research, scholars have proposed various features to describe specific classification objects, and texture features are one of the important statistical features. It reflects the distribution of the texture pixel structure space in the grayscale image, and is generally the repeated reproduction of a series of texture primitives in the pixel space according to certain arrangement rules. The research on texture feature expression method has great practical significance for content-based image retrieval, remote sensing im...

AI Technical Summary

Problems solved by technology

However, this three-value pattern is not strictly a three-value pattern, and its final encoding is still binary. Similar LTP variants include high-order LTP (high-order LTP) proposed by Zhang Y et al., these The encoding method can only be called "pseudo-ternary mode"

The OLTP proposed by Raja M et al., and the OS-LTP proposed by Huang M et al. both adopt the true three-valued mode coding, but they are not enough in terms of rotation invariance and feature dimension, which are exactly what texture features require. must-have key elements

[0006] The main disadvantage of the current local ternary pattern texture feature extraction method is that the dimension of the generated feature vector is too high to adapt to more than 16 neighborhood sampling points

Images