Study On Scene Recognition Algorithm Based On Fusion Features

Scene image analysis and recognition has become an important role in the field of image understanding, the main task is feature’s detection and description in the scene understanding,and get the semantic analysis and description of the scene. Scene recognition technology been used in the many fields, such as aerospace、the robot and biological recognition, and achieved good results. Good scene recognition technology can provide a amount of prior knowledge. In order to provide the prior information for object recognition field and other fields, this paper presents scene image recognition algorithm based on fusion features.This paper mainly completed the following work :(1)First, we summarize the technology and method in the field of scene image recognition both here and abroad, and analyze the problems in the scene recognition. we also summarize the method and technology in the feature extraction and classification which is the key to the scene recognition.(2)we import the GBD descriptor for local feature extract and propose the new HSV-GBD descriptor, the gradient of binary local descriptor Based on HSV color space and multi region.First of all, the new descriptor extract the gradient of different orientation which uses median difference, and reference the LBP algorithm to process the gradient image to become the final gradient image. Finally we use the histogram describe the image. We propose the HSV color space to extract the gradient image on each channel, at the same time, we make the image into uniform region and extract the local histogram and cascade them as the image local descriptor.(3)we propose a new comprehensive descriptor GIST-GBD, although the GIST descriptor can provide the global information of the image, at the same time it is coarse. In order to add the more local information of the image, we synthesize the GIST feature and the HSV-MRGBD descriptor and make the image has more information. First, we extract the global information of GIST feature and the local descriptor of HSV-MRGBD and fuse them to become the new feature of the image, finally we use the SVM classifier to classify andrecognize, and we compare the new method and the other methods, the experiment show that the accuracy is improved and the comprehensive performance also improved advisably.