Research On Natural Scene Image Classification Algorithm Based On Saliency Detection

Image classification technology is one of the most popular directions in computer vision research,and it is also the basis of some other image research application fields.It uses a computer to quantitatively analyze the image,and distinguishes different types of targets based on different features in the image to replace human interpretation.Compared with the more mature face recognition and other fields,images in natural scenes have higher requirements on the performance of image classification algorithms due to the complexity and variability of their environments and the diversity of object appearances.A large number of studies have found that the saliency detection algorithm can distinguish the background and the target object in the image by detecting the significant region containing the target object in the image.Combining it with a convolutional neural network can reduce the influence of the background region,thereby improving the performance of the classification.This paper first uses the saliency detection algorithm to detect the target region of the image,and then inputs the detection result into the GrabCut algorithm to achieve the automatic extraction of the target object.Finally,the extracted results are input into the convolutional neural network to complete the image classification.The specific work is as follows:1.In view of the problem that the background of the natural scene image is difficult to distinguish between the foreground region and the background,the superpixel segmentation algorithm is used to pre-segment the target image.SEEDS superpixel algorithm only use color feature of image to complete segmentation quickly,but it's not good enough for the segmentation of small color span or complex background natural image,the improved T-SEEDS algorithm is proposed by fusion the texture feature of image,and get a good segmentation results.2.For the interference caused by the complex background of natural images on the image processing,the image saliency detection algorithm is used to detect the target region of the image foreground,highlight the foreground and suppress the background,and reduce the influence of the background region in the processing process.In the saliency detection algorithm,the T-SEEDS superpixel algorithm is used to preprocess the image,which improves the efficiency of the algorithm and improves the accuracy of the detection.At the same time,for the problem that the saliency algorithm is insensitive to the objects on the boundary,the algorithm is improved by screening the extracted background seed points.3.Combined with the saliency detection algorithm and convolutional neural network,a new image classification model was implemented.The automatic segmentation of the image foreground is achieved by combining the saliency detection algorithm and the GrabCut image segmentation algorithm.Then the segmented results are input into the trained AlexNet convolutional neural network to complete the image classification.The foreground segmentation suppresses the influence of the background on the image classification to the greatest extent,and the convolutional neural network extracts the deeper features of the image with better adaptability.The final experiment proves that combining the saliency detection algorithm can greatly improve the performance of the image classification model.