Implementation of a Food Image Retrieval System Based on Color Feature Extraction Using Color Moments (CIELab) and Texture Feature Using Local Binary Pattern (LBP)

Abstract

The rapid growth of digital image content on social media has positioned food as one of the most frequently shared visual objects. Text-based search is often inadequate for retrieving food images due to their high visual complexity in color, shape, and texture, as variations in ingredients, lighting, and camera angle can produce highly diverse images even within the same food category. This study implements a Content-Based Image Retrieval (CBIR) system for food images by combining Grid Color Moments in the CIELab color space for color feature extraction and Multi-Scale Local Binary Pattern (LBP) for texture feature extraction. The CIELab color space was chosen for its stability under varying lighting conditions, while the multi-scale LBP approach captures texture patterns at several levels of observation. The system processes 3,600 food images across nine classes (Burger, Donut, Ice Cream, French Fries, Klepon, Fried Noodles, Fried Rice, Pizza, Rendang), using an 80:20 stratified split for database and query images. The retrieval process involves image preprocessing, feature extraction, feature vector normalization, similarity computation, and ranking of the most visually similar images. Similarity is computed using Euclidean Distance and Chi-Square Distance, evaluated with Precision@10 and Recall@10 on a Top-10 retrieval scheme. Results show the proposed feature combination improves performance over a baseline method that does not use grid-based and multi-scale feature extraction, with the best result achieved using Chi-Square Distance: Mean Precision@10 of 0.7133 and Mean Recall@10 of 0.0223. These findings confirm that combining color and texture features effectively represents food image characteristics and improves retrieval accuracy, offering a practical alternative to text-based food image search systems.