Font Size: a A A

The Research Of Real-time And Intelligent Image Recognition

Posted on:2013-08-06Degree:MasterType:Thesis
Country:ChinaCandidate:E M ZhouFull Text:PDF
GTID:2248330362473997Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
Image recognition is wildly welcomed by the industry for its non-contactcharacteristics and low-cost. With the development of image application, the recognition system is often designed to be very complex. The complexity of the systemextend the recognition time generally, this is fatal for the real-time requirementsoccasions, such as industrial detection and video surveillance.To resolve the conflicts between the performance of real-time and recognitionabilities, the thesis designed an optimized image system for the industrial productionenvironment.In order to improve the performance of real-time, the thesis improves a fast edgedetection algorithm for the industrial environment under simple background. The edgedetection time is reduced through the fan-shaped extension of the edge direction and theuse of historical information of the edge features. Secondly, the thesis defined acoefficient of variation which is used to filter the feature vectors. At last, the imagerecognition based on the improved ART2neural network which has a high speed ofcomputing, so, it makes sure that the system has a good performance of real-time.In order to improve the intelligence of recognition, the thesis improve the traditionalART2neural network, the improved ART2neural network is modified by being addedthe activities and adjustment mechanism of vigilance. It can show different learningpassion when it comes across image objects with different similarity under supervisedor unsupervised. It also can complete the study and recognition of image under themonitoring or non-monitoring.The experiment proves that the improved algorithm is efficient and useful. It is betterthan the traditional algorithm.The establishment of image recognition system is combined with the imagerecognition, image edge detection, feature extraction, matching organic and improvedART2neural network. In order to improve the flexibility of the system, the functions ofthe system is modular, each module complete a certain functions. The image recognitionsystem operation and test results show that the system works very well, so, the systemhas a high degree of practicality.
Keywords/Search Tags:Edge extraction, Feature vector, ART2neural network, Recognition, Activities
PDF Full Text Request
Related items