Design And Implementation Of Multi-object Recognition System Based On Multi-layered Structure

The research of machine vision based on optical images is vital and challenging.Computer and mechanical equipment can complate more extensive and complex tasks for human beings and the intelligent degree of society will be greatly improved, if machine vision can run as same as the human visual system.Generally, machine vision involves these steps: image preprocessing, image segmentation, object recognition and image analysis. Among them, image segmentation and object recognition are the most intractable and difficult problems. Most of the existing image recognition systems are based on image detection, and we can only deal with limited recognition problems with them.This thesis designs image processing algorithm and recognition system from image segmentation with the cue that machine vision mimics / simulates human visual, and achieves a multi-object recognition system based on multi-layered structure. The system uses multi-object segmentation achieved by a variety of image processing algorithms to extract objects instead of image detection. The object classifier consists of color layer,shape layer and local feature layer. It’s more close to the operational mechanism of the human visual.Finally, multiple sets of comparative experiments are carried out in this thesis. The experimental results indicate that the multi-object recognition system implemented in this thesis has the following characteristics: multi-object recognition, the system can recognize a variety of objects and the input objects are organized favorably by the system; automaticity, object recognition does not require additional manual operation;robustness, the system has the ability to handle large object database, and the multi-layered structure classifier is more effective than the single layered classifier;expansibility, the system can be reconstructed and expanded easily and deal with different classes objects recognition.