Visual identification of pointer instrument number has important significance to improve production efficiency of the electrical, mechanical, and automotive industry. In this paper, Hough transformation method was used in recognition of linear feature of pointer, and then dial circular feature was identified by an improved Hough transform. This work was completed by reducing the accumulated number of pixels and combining the gray center method. In the character recognition stage, BP neural network is built with a three level structure. Experimental results show that, this method achieves the expected effect for improving measures of Hough transform, and accurate recognition of characters can be obtained by using BP neural network.