False Alarm Filters in Neural Networks for Multiclass Object Detection


Authors: Mengjie Zhang, Bunna Ny
Source: GZipped PostScript (217kb); Adobe PDF (176kb)

This paper describes a neural network approach to multiclass object detection problems in which both the classes and locations of relatively small objects in large images must be determined. Rather than using high level domain specific features or raw image pixels, this approach uses low level pixel statistics as inputs to neural networks. The networks are trained by the back propagation algorithm on examples which have been cut out from the large images. The trained networks are then applied, in a moving window fashion, over the large images to detect the objects of interest. To reduce the false alarm objects detected, a false alarm filter is developed. This approach is examined and compared with a basic neural network approach on three object detection problems of increasing difficulty. The results suggest that the new approach with the false alarm filter can perform very well on those object detection tasks and is more effective than the basic approach. Keywords: Neural networks, pixel statistics, false alarm filters, multiclass object detection, object recognition, computer vision.

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