Textile Image Retrieval Based on BOF Approach

Content-based image retrieval (CBIR) has been one of the most vivid areas over the last 10 years. And how to retrieve images accurately and fast in a large database is extremely desired. However, different kinds of images have different characteristics, to better study the application of image retrieval, we take textile images for study. We presented the bag-of-features (BOF) approach, based on this, a plenty of local descriptors were extracted first. Here we propose a new approach for extracting the local descriptors, we combine together Harris corner detector and SIFT (Scale-invariant feature transform) method to generate the local descriptors, and we adopt the Locality-constrained Linear coding (LLC) to code the local descriptors into feature vector. For the images of the same category, we determine the similarity using color histogram and Histogram of Oriented Gradient (HOG) feature. We propose a new similarity matching algorithm which is more suitable for local feature matching. Experiments have shown this approach outperforms those retrieval methods which do not have a procedure of classification on textile image dataset.