A New Representation and Detection of Multi-Colored Object Based on Color Contents

摘要

Efficient content--based retrieval of complex images is a challenging task since the detected object may appear in various scale, rotation and orientation with a wide variety of background colors and forms. In this paper, we propose a novel representation of objects with multiple colors, the spatial neighborhood--adjacency graph(SNAG ). which can serve as a ba- sis for detecting object by color contents from the candidate im- age. The SNAG consists of a set of main-vertices and two sets of edges. Each main-vertex represents a single color region of multi- colored object, and edges are divided into two classes: Neighbor- hood edges representing neighborhood relationship between two main--vertices with similar color, and adjacency edges represent- ing adjacency relationship between a main-vertex and another vertex with different color. By investigating whether SNAG of object image is an isomorphic subgraph of SNAG of a candidate image. we can determine whether the similar object exists in the candidate image. In addition, we have also applied the proposed approach to a range of different object detection problems involv- ing complex background, and effectiveness has been proved.