A Domain Independent Approach to Multiclass 2D Object Detection Using Neural Networks and Genetic Algorithms


Authors: Mengjie Zhang, Victor Ciesielski
Source: GZipped PostScript (1131kb); Adobe PDF (1321kb)

This paper describes a domain independent approach to multiple class, translation and rotation invariant 2D object detection problems without any preprocessing, segmentation and specific feature extraction. The approach is based on learning/adaptive methods -- neural networks and genetic algorithms. Rather than using specific image features, raw image pixel values are used as inputs to the learning/adaptive systems. Five object detection methods have been developed and tested on three databases which represent detection problems of increasing difficulty. For detecting the objects in all classes of interest in the easy and the medium difficulty problems, a 100\% detection rate with no false alarms was achieved. The centred weight initialisation algorithm improved the detection performance over the basic approach on all three databases. In addition, refinement of weights with a genetic algorithm significantly improved detection performance on the three databases.

[Up to Computer Science Technical Report Archive: Home Page]