Research On Image Scene Classification Algorithm Based On Image Saliency

Due to the rapid development of multimedia technology,more and more image data is generated,so new technologies are needed to manage these huge image data sets.Scene classification is becoming a popular research field as an effective method for image understanding and management.In order to improve the efficiency of image scene classification,this paper integrates the visual attention mechanism into the image scene classification algorithm,from which the main work of this paper can be obtained as follows:(1)Aiming at the problem that the traditional background-based saliency detection algorithm can not accurately select the background seed points,which results in insufficient saliency detection results,an image visual saliency detection algorithm based on background seed points is proposed.The algorithm first performs weighted median filtering,SLIC superpixel segmentation,and Canny edge detection preprocessing on the image,then uses a minimum energy loss path algorithm to find more and accurate seed points that belong to the background,and finally based on the background seed A stream sorting algorithm is used to obtain the saliency map of the image.Experimental results show that the algorithm can improve the integrity and accuracy of the saliency detection area.(2)Aiming at the problem that the traditional visual word bag-based image scene classification algorithm does not distinguish between the background and the foreground of the image,which leads to insufficient classification accuracy,an image scene classification algorithm based on saliency detection and visual word bag is proposed.The algorithm first extracts SURF(accelerated robustness)local features from the image,then uses the method in(1)to calculate the visual saliency value of each feature point,and then fuse the extracted features with the visual saliency value to achieve the separation of the front and back of the image.For the purpose,a hierarchical clustering algorithm is used to obtain a visual dictionary,and finally the input image is represented as a histogram based on the visual dictionary,and a binary tree structured SVM multi-classifier is constructed for scene classification.The experimental results show that the improved visual bag-of-words model proposed in this chapter improves the average classification accuracy by 10% compared with the traditional visual bag-of-words model.(3)Aiming at the influence of background information of images on the classification accuracy and training cost of deep networks,an image scene classification algorithm based on saliency detection and deep learning is proposed.The algorithm first performs the algorithm of(1)on the input image to obtain the saliency detection result of the image,and then based on the saliency detection result,the foreground segmentation is performed on the original color image to obtain the foreground image,so that the background information is removed from the network The impact of the model.Then,a multi-scale residual neural network is used for feature selection of the foreground image,and the Softmax classifier is used to complete the final scene classification task based on the selected features.Experimental results show that the method proposed in this chapter can improve the classification accuracy by about 2% compared with the traditional convolutional neural network.