Research On Food Image Classification Algorithm Based SVM

Diet is one of the most important daily activities of human beings. With the gradual improvement of living standards, people are no longer satisfied to be able to eat belly, but began to pursue eat well, eat healthy. As the modern with high intensity of work pressure and fast-paced lifestyle, they will frequently intake of high calorie and high fat foods, which can cause high rates of obesity. Data shows that in the 21 st century, the development of obesity has plagued a global issues of country and society. However, a major cause of obesity is high intake of high-energy and high-fat foods. Therefore, more and more people began to pay attention to diet and physical health. If there is an image recognition algorithms to identify food image to help people record daily diet, this algorithm can not only monitor the daily intake of calories and nutrients, and can give a reasonable diet recommendations based on current diet and physical health. Therefore, we propose a simple and efficient food image recognition algorithm, not only has an extensive usage scenario, and with the slightest means of humanitarian concern.Image recognition is a hot topic in the field of imaging studies, it is widely used in face recognition, computer vision, robotics and many other fields. At present, among the domestic recent research results, we do not see much reaches on food image recognition. It's very difficult to recognition food images because of complex image background and image features.For this scenario to identify problems in the food image, we propose an ingenious solution. First, according to the generally color of the food, we divide the image into several color range indexing. This is because the food color is simple and fast to recognize, so that it can filter out many of the picture that significantly different from the food categories we want to identify. For example, most with vegetables as the main ingredients of the food, its main color is green, but most of the meat for food ingredients made, the main color is red. Secondly, in order to solve the complex problems of image background, we use Grab Cut preprocessing algorithm for image segmentation, extracted the food body we care about, weed out background pixel which will affect the classification accuracy rate. Experiments show that, compared with the FCM segmentation algorithm, Grabcut algorithm has obvious advantages in the food image segmentation. We then split out food target feature point extraction SURF, and use the bag of feature model extracted image bag of SURF feature. Finally, we use the extracted image features by SVM machine learning algorithms trained for each kind of food pictures, use one VS rest strategy to build food image classifier. In the design of the classifier, by extracting food image color index, you can filter out some of the identified distinct image classifiers to further enhance the performance of the algorithm.In the experiment, the method in use after GrabCut food image preprocessing algorithm libraries food image classification experiments, eventually reaching 71.3% of the top 5 recognition accuracy.