A Study Of Features Threshold Segmentation And Vector Of Graphics

Engineering drawing of automatic identification is an importantgraphics and image processing technique for the engineering drawingsrecognition system. But the previous engineering drawings are mostlydrawn by hands and they are fairly important. These engineeringdrawings are badly difficulty to maintain and index，and can’t be usedeasily for rear work. Engineering drawing recognition technique aims atconverting scanned engineering drawing images into vector formatscompatible with CAD systems，processing, analysis and identify theraster image and eventually converted into vector graphic format，tomodify and edit easily. The engineering drawings of automaticidentification can realize engineering drawing automatic inputeffectively. It can also set up a database in the computer graphics andshorten development cycle of CAD system.This paper systematically describes the structure，main algorithmsand their implementation of the offline recognition system for scanneddrawings, and studies the main components of the implementation of thealgorithm for engineering drawing scanned images, including imagesegmentation, image denoising, separation of text strings form graphics,image classification and graphics vector recognition of the digitalcharacters. Among them, some of the current image segmentationseveral threshold segmentation method are compared, and the iterativemethod is improved; Image denoising part is using adaptive filtermethod to eliminate the noise of the binary image; The image of refinedrefinement of demand is higher, the traditional Hilditch method easy toproduce the distortion, the index table refining method in this paper canavoid this situation; The vector uses method combined of based ondetailed methods with adaptive grid, can avoid the distortion of thecrossing of the traditional based on thinning method. In this paper, thecharacter recognition was also studied, mainly on the number recognition of the engineering picture, designing a digital identificationsystem by improving the BP neural network.In this paper, some algorithms of engineering graph automaticrecognition have been improved, and some experiments and analysishave been done, and has some academic and applied values to thetechnology of recognition and understanding for scanned engineeringdrawings.