Feature extraction and pattern classification are both important issues in pattern recognition fields, among which, texture analysis is one of the basic problems in image processing and computer vision. Based on the newly proposed fabric texture analysis method via constructing explanatory variables with associated dynamic evolutionary complex networks, we carried on the study of designing of strategy on high-dimensional feature descriptor vector’s generation and feature extraction process of texture images on details. The texture signature vector of given images was constructed based on this method, which can effectively characterize the digital features of fabric texture images. This method was successfully used to classify Kylberg texture image libraries according to the numerical experiments. Comparison numerical results via different classical classification methods were also carried out and the related results showed that this method was not only rotation invariance, but also with robustness against noise. It might have potential industrial application prospects.