Research On Automatical Collection And Recognizetion Methods Of Test Paper Information

Marking system can not mark handwritten content on test paper. The basic reason is that the recognition technology can not achieve an ideal recognition rate. Considering of this condition, examinee's name, class and mark are collected as samples to recognize. Based on some previous algorithm, theories and improved algorithm, we execute these experiments. System includes collection of test paper image, segmentation of interesting area and recognition of handwritten character.The collecting work is collection of static test paper image. In the process of the collection, we completed the job by using professional image collection instrument and adopting device's function of continual frame. Then, we select these samples by using manual method. Partial automation is obtained by hard work.Segmentation of interesting area achieves aim of segmentation of test paper header and handwritten Chinese character. There are a lot of methods to Segment the header of test paper like examining straight line algorithm and orientation of segmentation region algorithm. Considering RGB color space is sensitive to light, we choose technique of HSV color space segmentation. In the process of segmentation, we use average value of 8-area way to complete segmentation. Experimental data proof that above 90 percentage of the experiment succeed.Pretreatment includes image enhancement, removing noise, image binary, normalization and thinning. Given the collection of flexible environment, we enhance the image before binarizing it. Experimental data show that the means we adopted obtain an excellent binary result. We use wavelet transform and elastic grid technique in Extraction feature phase. This means can solve problem that single feature can not achieve high recognition rate. In classification phase, we choose BP neural network. Hidden layer is changed depend on convergence rate. Experiment data illustrate that BP neural network can obtain high recognition rate in digital character and Chinese character recognition experiment.