Segmentation Method of Boxed Lunch Images by Weight Optimization Using Color Texture Dissimilarity

People with severe upper limb disabilities need assistance from caregivers when eating, which becomes a burden for caregivers. Our laboratory has developed meal assistance robots that use image recognition and voice recognition to solve this problem. In a previous study, we proposed a region segmentation method using color and texture differences derived from input images as an image recognition method. However, this method was less effective because lighting and shadows affected the selection of appropriate features in the calculation of differences. In this study, we introduce a method for calculating the degree of difference between color segmented regions based on the HSV color space and texture obtained from color information. Furthermore, we propose a method that uses the HSV color space based on saturation and texture weight optimization to calculate differences to perform environment-independent region segmentation. A validation experiment was conducted to properly divide the food area of boxed lunch to confirm the effectiveness of the proposed method. The results showed that the accuracy exceeded 90% for all boxed lunch. Therefore, the proposed method can accurately identify food areas, regardless of the environment.