Shape And Texture Based Scene Classification

Humans are extremely proficient at perceiving natural scenes and understanding their contents. Scene recognition in Human is the natural activity by which human can easily recognize the scene even if the scene is complex, partially occluded or blurred. In machine vision the recognition rate is less compared with human vision. To improve the recognition rate of the machine vision an efficient structural and textural based features are extracted from the image. H-Descriptor with Local Binary Pattern (LBP) [24] and H-Descriptor with Local Gradient Pattern (LGP) can effectively extract structural arrangement and textural arrangement of pixels in an image. LGP is invariant to local intensity variation so it is efficient for scene classification. LBP and LGP [23] is applied for each slices when the input image is separated into three different slices. Then Haar wavelet is applied for the input image and three different slices. The HOG is applied for each Haar wavelet transformed images to produce H-Descriptor with Local Binary Pattern and H-Descriptor with Local Gradient Pattern. Then by taking the H-Descriptor with Local Binary Pattern and H-Descriptor with Local Gradient Pattern as two independent feature channels, and combined them to arrive at a final decision using SphereSVM [22] for achieving an effective scene categorization. Keywords— H-Descriptor; Local Binary Pattern; Local Gradient Pattern; Haar wavelet; SphereSVM