Novel approach to object recognition by the fusion method

摘要

Abstract: recognition depends mainly on extracting theoptimal features which should be insensitive to imagetranslation, scaling, rotation, and noise. However, itis a complicated and difficult process, and stablefeatures may not be extracted in some cases. In thispaper, a novel approach to object recognition based ona single sensor is proposed in the view of data fusionand Dempster-Shafer's theory. The notions of imagesubfeature and similar degree function (SDF) are firstintroduced. For each SDF function, we further establisha set of subordinate functions (SF). The SDF and SF arecombined in the fusion model. For each class oftraining samples, several subfeatures are selected bythe different methods. Then, the SDF and a set of SFfunctions are calculated. Finally, the Dempster's ruleof combination is used and all subfeatures are fused.In the fusion model, a simple classifier is designed torecognize objects. Experimental results show that theproposed method is efficient and our recognition modelhas good performance. !5