Image Content Descriptor Based Scene Classification Research

The process of scene image classification is an important branch of image understanding, which using belongs to the part of image classification. It makes computer understanding the overall image semantics and automatic marking label image scene. Complete the scene classification needs to extracting the image features, forming the difference between different categories of content based image description, and through the classifier to complete the sentence. This paper introduced image content descriptor based scene classification research.First of all, the traditional way of meshing rules are likely to be that the same target points to different grids, and the pixel grid can split along the edge of target segmentation, ensure the integrity of their goals as much as possible. Different scenes’ space change direction is different. In order to highlight the importance of this kind of space structure, extracting horizontal and vertical gradient of different images can obtain similar the weighted coefficient of vertical and horizontal. Weighted on super pixel up-down or left-right grids can describe the content of image.Secondly, local descriptors are suit for scenes which locally differences are huge, while global descriptors are suit for scenes which globally differences are huge. Extracting local features, adopting different ways of mapping and increasing super pixel space pyramid grid to describe the image. Specially super pixel grid makes the image target as complete as possible and features in the image block as consistent as possible. Then global features and local features are combined, and makes the image description includes both the structure of local information and global information.Finally, the indoor scene has higher similarity between classes and bigger difference within class, the classification of indoor scenes is more difficult than the natural scene images. Scene especially indoor scene image contains many objects. Describing the objects in the image firstly, then represent the scene may help to complete the scene image classification. Salient grids partition and spatial grids partition get salient information and structural information of image at the same time. With object detectors to characterize different areas of the image will complete indoor scene image classification.