Research On High Accuracy Recognition And Preprocessing For Handwritten Character

Handwritten character recognition technology is an important issue in the field of human-computer interaction and in recent years has got rapid development. Its applications cover a wide range, involving civil and military. Such as electronic blackboard teaching, which can identify numbers and formulas online; It can also be used to draw and identify electronic components, for the identification and recognition of the military road signs and other labels. All of these applications require a high recognition accuracy and fast recognition speed, so handwriting character recognition require a high recognition rate and need to be real-time. This paper proposes a method that can be used to identify the handwritten symbols in the election vote recount and achieved high recognition accuracy and speed.Commonly used special handwritten symbols inside ballot image include tick ("√"), circle ("○"), fork ("×"), bars ("\,-,/" three kinds of writing). The proposed symbol handwriting recognition method can solve quickly locate problem electoral in counting system with a handwritten symbol recognition based on OCR(Optical Character Recognition) technology. This method not only statistics information of the ballot, but also can realize the visual design. The handwritten character recognition method is divided into preprocessing stage and symbol recognition stage.The preprocessing stage is to locate the position of handwritten symbols. First, we use a scanner to transfer images, replacing high-definition camera to collect data, and then use a TWAIN technology and workflows to access to the ballot image, which ensure the integrity and reliability of the data. Then we do the gray processing for ballot images. The original image bright range in each RGB component is 0 to 255, so the RGB color space has R* G* B=16581375 kinds of color. Will be very troublesome for each RGB component range in the original image are 0 to 255, so that the RGB color space can be expressed as R* G* B=16581375 single color, in this case of great amount of data, processing will be quite troublesome and time consuming. We let the three RGB component values equal, and use one component to represent, thus reduce the amount of raw data. Then we do the ballot image binarization. The so-called binarization is that, we set a threshold, the value of pixels in the image set to white if it is smaller than the threshold, otherwise set to black. Then we do the ballot image denoise. The reality of the digital image affected by devices or the external environment such as noise in the digitization and the transmission process, the image usually contains noise, i.e. redundant data interfere people to understand or acquire image information. Denoise is to reduce the redundancy data. The final step in preprocessing is line detection and table localization. This step is to identify the precise location of handwritten symbols in order to get a handwritten sign at the time of symbol recognition.In symbolic recognition phase, the incoming parameter is the image and the positions of handwritten notations; the image has been through the gray scaling, binarization and denoises, but also need to do image deformity correction. Four corners are passed to function to represent position, after scaling transform, the position becomes horizontal and vertical rectangle. Then we analyze base structure element features for symbols, and classify these handwritten symbols. Handwritten character recognition also includes a rectangular code and bar code recognition. Rectangular codes and barcodes can be used to determine the back or front side of image, up and down direction, the electoral vote category and other special-purpose. Because the direction of the image is unclear after scanning, so it is important to determine front or back side and up or down direction. Because handwritten character recognition method in this paper can be used for election recount, it is a good idea to classify electoral votes with rectangular codes or bar codes. When rectangular codes or barcode is used to distinguish voters, we statistic election difference between different kind voter; when used to identify a ballot type, it is not a illegal ballot if scanned image code value is not the same with we set. On the one hand, this can reduce the statistical errors with invalid ballot, on the other hand distinguish different election categories.