A Multi-Neural Network Approach to Image Detection and Segmentation of Gas Meter Counter

In this work we propose an automatic approach to detect and recognize text in natural images. Differently from other techniques which extract text from the whole image, our approach is trained to detect a particular area of interest where performs the search of text. Firstly we detect the region of interest and then we detect and segment the characters found. The proposed method may be used to read a generic text contained in a specific object from a single image. To show the results of our approach we focused on the problem of the automatic gas meter reading. The goal is to process a natural image containing a gas meter in order to extract only the information from the digits of the meter counter. Despite the complexity of the problem, our approach shows good results and the low computational time required by the algorithm permits to use it in real applications and in mobile devices.