Study on the Vision Reading Algorithm based on Template Matching and Neural Network

An improved solution to the vision reading of the traditional UTM (universal tools microscope) is introduced. Based on the comprehensive analysis of the current working algorithms, a compound vision reading model is proposed after adopting several pre-processing algorithms. This model is established with the template matching method and BP neural network technology. In order to improve fault-tolerance capacity of the network, rotation invariant features based on Tchebichef moments are extracted from numeric characters and a 4-dimensional group of the outline features is also obtained. The experimental result shows in applying the newly developed algorithm, the measurement accuracy of the automatic vision reading can achieve plusmn3 mum, while it may reach or surpass ocular reading precision if the mm reticle is overlapped with the drum wheel.