Research On Character Detection And Recognition Method In Engineering Drawing

With the advancement of science and technology and the development of computers,industrial production has gradually been transformed from worker-led production and testing into automated processing by computers.In the process of designing,manufacturing and testing mechanical parts,using computers to detect and identify instrument screens and drawings Characters,and automated analysis and process of these parameters,without the need for manual testing and comparison of drawings can greatly improve the production efficiency,while reducing the error caused by human beings.In this thesis,based on the production and testing of mechanical parts,the key step in the entire automated quality inspection is completed by studying the detection and recognition methods of characters in the instrument and drawing images,and a complete system of character recognition on instruments and drawings are designed and implemented.The research contents of this thesis are mainly as follows:1.The character detection related algorithms are studied.According to the fact that there are many marking lines,graphics of instrument and machine component,indicator symbols and other disturbances in the image of instruments and drawings,an improved character detection method is proposed.Through a degree of image processing and some filter conditions,these disturbances are removed,and then combined with the detection algorithm to do a rough extract,finally,based on an algorithm of connected domain aggregation,a series of required target text lines are extracted.2.The tilt correction and character segmentation of text lines related algorithms are studied.In view of the poor performance of traditional correction algorithms in the actual situation in industrial production,a relatively effective tilt correction algorithm is adopted,and an improved segmentation algorithm of overlapping characters is proposed for complex situations.The experiment show that this character segmentation algorithm is obviously superior to the traditional methods,and can effectively complete the correction and segmentation.3.The algorithms of feature extraction and classification in character recognition are studied.The various features and classification algorithms are tested for the recognition of characters in actual projects.For the fact that the invariant moment features are not performed effectively,a corresponding feature selection and correction are carried out.And for the shortcomings in the KNN classification algorithm,an improvement based on distance weighting in the final classification decision stage is proposed.Then through experimental verification,the improved feature extraction and classification algorithms have significantly improved the recognition of instrument and drawing image characters.4.The errors of recognition of the instrument and drawing image characters are mainly analyzed,and the errors are concentrated on the segmentation methods and the recognition of similar characters errors.For segmentation errors,structural features and recognition feedback are used to improve the segmentation accuracy.While for the recognition of similar characters errors,a recognition confidence and context-based correction mode are proposed to improve the accuracy.Finally,through a combination of multi-level recognition modules,a more effective recognition process is designed to further improve recognition rate and reliability.The algorithm module involved is programmed,tested and integrated to complete the design and implementation of the whole system.Through the analysis of the test results,the algorithm and system designed in this thesis can effectively meet the actual production requirements and achieve the expected goals,thus it can be applied to industrial production to improve production and test efficiency.