The Research On The Technology Of QR Code Recognition

In our country two-dimensional Bar Code has been widely used in many industry. To shoot two-dimensional Bar Code with universal shooting equipment and recognize it is one of the important trend on two-dimensional Bar Code's development.In the article coding and decoding QR Code is discussed. Aiming at blurry QR code image which is captured by camera, a application scheme of examinee identity authentication using QR Code is put forward.conventional two-dimensional Bar Code's recognition with image technology is contrasted and researched.Conventional method departs code image from originally captured image using Hough transform and edge examination, adaptively chooses threshold to wipe off inefficative borderline caused by noise using Flourier transform, to get basic module of code, but its inspect efficiency to defocused image is not stable, and the complexity of Flourier transform is somewhat high. The article uses median filtering, binarization, image orientation, image rotation, image partition and decoding to get recognition. Major innovation is improving median filtering to decrease noise and choosing dots with beeline before Hough transform.In frondose median filtering method, the article combines the excellence of extremum median filtering and self-adaptive median filtering, to improve on classical self-adaptive median filtering. Major antilogy is: departing all pels into doubtful noise and signal to dispose secondly for decreasing miscarriage of justice, and disposing differently on low-density area and high-density area to relax the conflict of protecting detail and filtering noise.Checking doubtful pels secondly with second checking-window improves veracity of noise-detect, and won't cause too much affect to calculating speed. The pels which can be found median in layer one and is detected as noise, are filtered according to texture tend using eight representative texture child-window. The pels which can be found median in low-density area can be filtered considering local detail. The pels in high-density area which still can't be found median in layer one when the window size is max, are detected secondly if they are judged as noise in layer two. If they are judged as signal then they are output, if they are still judged as noise then their curve imitate join are output.One step is added before Hough transform, that is drawing line in horizontal and vertical direction, noting the first pel every line meets, abandoning others. Latter Hough transform can be disposed only to those pels, the calculating quantity is much reduced.