Algorithm Of Target Recognition Based On Moment Invariants

The moment technique is widely used due to its appealing mathematical simplicity and versatility, such as image processing, computer vision and pattern recognition. According to the basic principle of image segmentation, a pre-processing algorithm is showed, which is fit for this thesis. Aim at the target which has been translated, scaled and rotated, the paper shows a method which uses moment invariants as the matching feature and compares the performance of Hu's moments based on region and Chen's methods based on region boundaries using two badminton images, the experimental results show that Hu's moments are more steady and valid. Finally it realizes the target recognition of 2-demision badminton with Hu's moment invariants and gets a comparatively ideal result. This paper offers an effective method for target recognition with random shape in complex condition and has practical evaluation.