Supermarket Goods Recognition Based On Image Processing

As an essential shopping place in modern society, supermarket becomes more and more favored by customers. Goods are the core of the supermarket, the accuracy and real-time capability of the shelf goods information obtainment will directly affect the operation efficiency and comprehensive competition of the supermarket. At present, these tasks are completed by a large number of staffs and this method has several defects such as low working efficiency, heavy workload and so on. If we can develop an intelligent management software which can automatically recognize the goods on shelves, it would be more timely and accurately to obtain the information of goods, and much helpful for supermarket managers to make reasonable decisions.According to these disadvantages, this paper designs a new method for supermarket goods recognition, which is based on image processing. Goods recognition includes two parts, namely image feature extraction and pattern recognition. We mainly use the statistical features. Firstly, extract features of gray goods image by Principal Component Analysis(PCA) method and Linear Discriminate Analysis(LDA) method, and further verify and compare these features experimentally. Secondly, considering the color information of the images, we extract features of color components in RGB space and HSI space respectively, and then design three classification decision criterions to recognize goods, which make the recognition results much better than only using gray features. Lastly, recognize goods by using the method called two-pass classification based on HSI color features. This method combines the Hyper-Ellipsoid Neural Network(HENN) and the error correction SVM(EC-SVM) according to their outstanding characteristics, and obtains quite better recognition results. Simulation result shows that with the method of combining LDA features and two-pass classifier, the recognition rate reaches99.5%in the experiments on goods database, and97.07%in the experiments on shelf goods recognition. These results verify the effectiveness of the goods recognition method this paper proposed.