Abstract: This paper proposes a method for recognizing multi-food images by detecting foods in candidate regions and classifying them according to different characteristics.The candidate areas are detected by fusing the output results of multiple area detectors to combine the image feature MKL of each category to estimate the optimal weight.The food recognition method based on feature fusion was applied to the bounding box of candidate regions with various visual features,and through experiments,arranged in descending order of score to estimate images of multiple candidate foods.The results show that for a variety of food image data sets,a classification rate of 55.8% is achieved.The research results have certain reference significance for the field of similar food identification.

Key words: multiple food images, area detection, window search, multi-core learning