Fast Object Classification Method Based on Saliency Detection
Yang Chao, Wang Zhennan, JIANG Xuepu, Yu Hongyan, Sai Liu, Deng Liangzhu · 2018
In recent years, image classification has become a research hot area in computer vision, Most of the methods directly train data sets to obtain classifiers. However, if the images in the data set contain massive backgrounds, causing the image itself to have a lot of noise will degrade the classification results. This paper presents a new object classification method based on salient object detection. Firstly, a salient object detection method based on CNN is applied to the dataset image to locate the object in the image. Then use the Bag of Feature model to vectorize the salient region features. Finally, use the multi-class linear SVM to classify the feature vectors. For marveldataset2016, the classification accuracy of the pre-processed classification method is 10.8% higher than that of the non-preprocessing classification method, and the classifier-training is 44% less time-consuming.