Study On The Embossed Character Recognition System Based On Neural Network

The embossed character is widely used as a marked character. It is made by amold pressing on the surface of the object so that the surface is deformed to form aconcave-convex character. So, the material of the character is same to thenon-character region. When camera is applied to acquire embossed character image,the characters become illegible. In addition, gaps between the character molds inducethe embossed features in the non-character region. Therefore, it is not able to meet therequirements of embossed character recognition by directly use existingcharacters algorithm. With the development of the modern industrial technology, theembossed character recognition technology has become an important part of themanagement of product information. The embossed character recognition technologyis thoroughly studied in this paper and carried out the following researches:Study on the machine vision system for acquiring image of the embossedcharacters on the tail of the toothpaste. For the embossed character recognition, theacquisition of the clear character image directly affects to the final performance of theembossed character recognition. This paper first introduces the composition of themachine vision systems, machine vision systems hardware parameters and theirselection principles. Then select the appropriate hardware to construct the visualsystem requirements. And according to the characteristics of embossed characters, it isdesigned as visual system of the character recognition system that uses low-anglelight source lighting source with frosted glass. Finally, the effectiveness of the visualsystem is verified by toothpaste with single-row and double-row characters. Thisvisual system can achieve strong contrast between the characters and background ofthe embossed character in the image and has laid a good foundation for thepost-image processing algorithms.Study was also delivered on the location algorithm for embossed characters base ontemplate matching and the Otsu segmentation method for single character. This paperfirst introduces the character location method based on template matching based onthe characteristics of the embossed characters and known information. In addition,since the deformation of the character and the overlap caused by gaps between wordmolds must be taken into account, it is mandatory to put forward a precise embossedcharacters location algorithm based on template matching method. It comprises two steps: firstly coarsely locating the region of interest for multi-character charactersbased on template matching algorithm, then accurate find the precise positions of thecharacters in the region of interest using a template matching algorithm. From theexperiment, the algorithm can achieve precise finding the embossed characters. Afteraccurate location of the embossed characters, the single character segments resultusing the Otsu method was finally well obtained.A combined character recognition algorithm of BP neural network and characterstructure characteristics were proposed. It solves the error reorganization of somecharacters and achieved very perfect performance. In our experiment, the characterrecognition system to identify the correct rate of system design requirements, toachieve a good recognition results.