Image Retrieval System Based on Machine Learning and Using Color Features

We describe an interactive system for content based image retrieval. The system presents the user with 15 randomly selected images from the database. The user grades the images with one of five possible grades (YES, yes, neutral, no, NO) according to what he is looking for. The system returns the first 15 images with the highest probability of YES grade. The attributes used are a combination of color features. Three different machine learning techniques are compared.

• Type: Conference
• Year: 1999
• Where: CAIP
• Authors: Janez Demsar, Dragan Radolovic, Franc Solina
• Keywords: CAIP 1999, Content Based Image Retrieval, Image Analysis, Images, first 15 Images