AI Summary & Key Takeaways
AI InsightExperimental results show that our proposed descriptor yields good classification accuracies (the best results on Coil-100, and the second best result on Caltech-101 and PASCAL VOC 2007 datasets) indicating that FT based local descriptors obtain important properties of images that are valuable for visual object classification.
📌 3 Key Takeaways
Abstract
Image representation models such as bag of words (BoW) or Fisher vector (FV), which are built depending on encoding local features, are commonly used in visual object classification tasks. In this context, the local patches sampled from images are represented by different texture and shape descriptors such as Scale Invariant Feature Transform (SIFT), Local Binary Pattern (LBP), Speed up Robust Features (SURF), etc. In this study, we propose a new descriptor using weighted histograms of phase angles of local 2-D discrete Fourier transform (FT). We make comparisons with the classification accuracies achieved by using the proposed descriptor to the ones achieved by other commonly used descriptors on Caltech-101, Coil100 and PASCAL VOC 2007 databases. Experimental results show that our proposed descriptor yields good classification accuracies (the best results on Coil-100, and the second best result on Caltech-101 and PASCAL VOC 2007 datasets) indicating that FT based local descriptors obtain important properties of images that are valuable for visual object classification. By combining image representations resulting from FT descriptor with the representations resulting from other descriptors, accuracies even get better demonstrating that tested descriptors encode different supplementary knowledge.