A Texture Feature Extraction Method Based On Frequent Itemsets And Its Application In Image Classification

At present more and more information is transferred and used in the form of images, and there is a demand for advanced technologies in image processing and analysis. As one of low-level features, texture describes important information of an image. And thus to effectively analyze and extract texture information from an image is very important in the task of image retrieval and classification.Wavelet transform has good time-frequency and space-frequency localization characteristics, and the coefficients of an image represented by the wavelet can still retain original local details. This paper studies the extraction method of texture features based on frequent itemsets in wavelet transform domain and expounds how to use the method in the process of texture classification. Texture features obtained by our method and color features are then combined to classify color images for better performance.With intensive study of wavelet and data mining, we propose a new method of texture feature extraction based on frequent itemsets, which employs the combination of frequent2-itemsets and statistical features as texture features. DWT is firstly utilized to decompose images into different scale subbands, from which texture features are to be extracted. Frequently occurring local structures in images are captured from the approximation subband of one-level DWT decomposed images in the form of frequent2-itemsets, which contain both structural and statistical information. Statistical features of the detail subbands are then calculated and combined with the above features to distinguish texture classes. The experimental results verify the good performance and suitability of this method.Image classification only using single texture feature often lacks capacity for representing the whole information of an image. This paper incorporates the texture feature and color histogram to classify color images. Experimental results show that using the combined feature can obtain better classification results than using a single feature.