A Domain Independent Approach to 2d Object Detection Based on Neural Networks and Genetic Paradigms (2000) [5 citations — 1 self]
|
Abstract:
The development of traditional object detection systems usually involves a time consuming investigation of good preprocessing and ltering methods and a hand-crafting of dierent programs for the extraction and selection of important image features in dierent problem domains. To avoid these problems, this thesis describes a domain independent approach to multiple class, translation and rotation invariant object detection problems without any preprocessing, segmentation and specic feature extraction. The approach is based on learning/adaptive methods { neural networks, genetic algorithms and genetic programming. Rather than using specic image features, raw image pixel values or pixel statistics are used as inputs to the learning/adaptive systems. Six object detection methods have been developed and tested. These are (1) the basic approach which uses multilayer feed forward networks trained by the back propagation algorithm, (2) a centred weight initialisation method which improves the performance of the basic method, (3) a method which uses a genetic algorithm to train neural networks, (4) a method which uses a genetic algorithm to rene network weights obtained in the method of the genetic algorithm
Citations
| 14 | Genetic programming for multiple class object detection – Zhang, Ciesielski |
| 2 | Using back propagation algorithm and genetic algorithms to train and re neural networks for object detection – Zhang, Ciesielski - 1999 |
| 1 | Centred weight initialisation to improve the performance of network training speed and the performance of object detection – Zhang, Ciesielski - 1998 |
| 1 | Centred weight initialisation in neural networks for object detection – Zhang, Ciesielski - 1999 |
