| Summary: | 碩士 === 國立交通大學 === 光電科技學程 === 105 === In recent years, smart phone application software has developed more and more rapidly, the species is also quite diverse, has been integrated into one of the main tools of life and leisure. This case is mainly to develop the App image recognition software to test the answer film identification rate, the experimental conditions must be completed 100 test answers to identify the success rate of 100% to achieve the goal, the test process to test how to overcome the encountered Questions, such as: the answer card tilt, writing too scrawled, not written in accordance with the provisions of the format, line segmentation of the calculation of shortcomings, thin line processing and the expected results do not match, noise interference. These problems belong to the non-consistency caused by the identification of abnormal, to take the image recognition software to improve is only a certain effect, and can not be completely overcome. Therefore, this study will set the Auto off-set in the image recognition, reduce the problem gap and automatically determine the error to improve the image recognition rate, but there will be time to delay the phenomenon, resulting in the entire identification process time, or even interrupt the phenomenon, in order to avoid the final The output of the wrong result, this factor is acceptable. The main features of the system are: (1). Acceptable non-A4 format photocopy paper, non-traditional answer card form (2). Acceptable non-2B pencil painted color, color contrast can be identified (3). Identify changes in the angle of
shooting up and down.
|
|---|
