Scene image recognition based on the sequence of local image vectors represented by oriented edges

A new feature description algorithm has been developed for scene image recognition. Since scene images have a large variation in the same scene category, the sequence representation using local image vectors has been explored in this work. The structural features present in scene images are enhanced by using oriented edges to represent local images. As a result, it has become possible to handle the large variation in scene images robustly. Using the test images from seven categories, extensive experiments were carried out. In some of the sample images, test images were resized by either changing the resolution or cutting off at the peripheral area, and evaluated the robustness of the algorithm. The results were compared to Scale Invariant Feature Transform (SIFT) [1], and robustness of the proposed method against the variation in scene images has been confirmed.