Study On Plate Lattice Spray Printing Character Recognition Method

In order to detect size of plates in real-time, lattice spray printing characters on plates are needed to recognize. Taking recognition of lattice spray printing characters on plates in a steel plant as the research background, this paper proposes a recognition solution. The main research contents are as follows:To solve the problem of strong noise and uneven brightness in images of lattice spray printing characters on plates, Gaussian filter and median filter are adopted to remove image noise; Top-hat transformation is used to eliminate image background of uneven brightness; Otsu threshold algorithm is used to process image binarization of characters on plates.For the segmentation issue of lattice spray printing characters on plates, the projection method is adopted to process positioning and segmentation of lattice spray printing characters on plates using prior information such as width and height of characters.For the recognition issue of lattice spray printing characters on plates, the paper proposes a character recognition method based on weighted feature template matching due to low recognition rate in traditional template matching. Stroke feature of lattice spray printing characters on plates is extracted, which is assigned to different weight: pixels located in the stroke center have highest weight and pixels located in the stroke edge have lowest weight. Weight of pixels at any position in character image is determined by pixels in eight positions surrounding pixels to recognize weighted feature character image of lattice spray printing characters on plates.For the recognition issue of lattice spray printing characters on plates, the recognition methods based on BP neural network and probabilistic neural network are also adopted in this paper. A method of per pixel feature extraction is adopted in the recognition method based on BP neural network; feature extraction of character image using structure characteristics and statistical characteristics method are adopted in the recognition method based on probabilistic neural network.