Learning Features for Object Recognition

Abstract

Abstract. Features represent the characteristics of objects and selecting or synthesizing effective composite features are the key factors to the performance of object recognition. In this paper, we propose a co-evolutionary genetic programming (CGP) approach to learn composite features for object recognition. The motivation for using CGP is to overcome the limitations of human experts who consider only a small number of conventional combinations of primitive features during synthesis. On the other hand, CGP can try a very large number of unconventional combinations and these unconventional combinations may yield exceptionally good results in some cases. Our experimental results with real synthetic aperture radar (SAR) images show that CGP can learn good composite features. We show results to distinguish objects from clutter and to distinguish objects that belong to several classes.

Automatic Design and Synthesis of Automatic Target Recognition (ATR) Systems Using Learning Paradigms

“…The experiments show that if training regions are carefully selected from a training image, the best composite operator generated by GP is effective. In the following experiments in sections 1.3.1 and 1.3.2, the parameters are: population size (100), the number of generations (70), the fitness threshold value (1.0), the crossover rate (0.6), the mutation rate (0.05), the maximum size of composite operator (30), and the segmentation threshold (0). In each experiment, GP is invoked ten times with the same parameters and the same training region(s).…”

“…The results from the run in which GP finds the best composite operator among the best composite operators found in all ten runs are reported. The parameters are: population size (100), the number of generation (70), the goodness threshold value (1.0), the crossover rate (0.6), the mutation rate (0.05), and the segmentation threshold (0). The GP program ran on a Sun Ultra 2 workstation.…”

“…Unlike the work of Stanhope and Daida, the primitive operators in this chapter are not logical operators, but operators on real numbers and the composite operators are binary trees of primitive operators on real numbers, not binary trees of logical operators [70]. They use GP to evolve logical expressions and the final outcome of the logical expression determines the type of the object under consideration (for example, 1 means target and 0 means clutter).…”

Discovery of Human-Competitive Image Texture Feature Extraction Programs Using Genetic Programming

“…Ross et al [10] used statistical features to classify microscopic images of minerals. Lin and Bhanu [11] used a co-evolutionary approach to build composite features from primitive features. This approach involves the steps of training and classification shown in Figure 1.…”

Combining Adaboost learning and evolutionary search to select features for real-time object detection

“…Krawiec [8] extends standard GP by a local search operation for visual learning. Lin et al [9] propose a co-evolutionary GP to learn composite features based on primitive features that are designed by human experts. Bala et al [6] combine a Genetic Algorithm (GA) with decision tree learning: The GA selects a good subset of features from a fixed set and a decision tree is learned to build the detector structure.…”