3D object recognition using multiple features and neural network

Abstract

To improve the performance of view-based three-dimensional object recognition system, we propose to extract multiple features from the 2D images of 3D objects, including texture characteristics, color moments, Hupsilas moment invariants, and affine moment invariants. Texture characteristics and color moments are used to distinguish objects of similar shape and different appearance. Hupsilas moment invariants have the invariance properties under rotation, scale and translation, and affine moment invariants have the invariance properties under affine transformation for the 3D objects in images. All these characteristics compose a 1-dimensional feature vector of 23 components for each 2D image of 3D objects, and then they are presented to a BP neural network for training. The trained BP network can be used to recognize 3D objects when provided the feature vectors of unseen views. We assessed our method based on both the original and noise corrupted COIL-100 3D objects dataset and achieved 100% correct rate of recognition when training views were presented every 10 degrees.