Automatic region of interest extraction in food baking images

Abstract

This paper presents a novel approach to automatically detect the region of interest (ROI) in food baking images. In addition to common ROI extraction and saliency extraction problems, food baking images are also subject to lighting noise, glass reflection and shadow effects. In this paper, the HSV color map is employed to effectively eliminate oven lighting effects, then a one-bit transform (1BT) with global adaptive thresholding method is performed to obtain preliminary ROI segmentation results. Further, a local contour refinement algorithm is designed to reduce glass reflection and shadow effects for extracting the ROI contour more precisely. The ROI extraction results presented in this paper can be used in compression and transmission of food baking images and videos for further analysis.