ABSTRACT
Feature extraction and identification for raised or indented characters pressed on metal labels is researched still less, Li Guoping [1] reflects the excellent characteristics by using Canny edge detection, with its characters as a convex edge extraction tool, then filled with the characters, using the projection method to preliminary segment character images that had been filled, and then refilled each images. Now researches for characters bump are rarely at home and abroad, Cao [2] has reported, his feature extraction of raised or indented characters using circular projection method to obtain a one-dimensional character features, and then wavelet, DCT transformation method, such as the combination of the rough character recognition or detection has a better result. This feature extraction feature extraction method with gray, scale, and rotation invariant, and good noise immunity. In recent years, due to the insensitivity of directions, scales and illumination, histograms of oriented gradients (HOGs) has received a wide application. The method of HOG was initially put forward by Navneet Dalal and Bill Triggs [3] in their article. This method was mainly applied to the pedestrian detection in static images, but later, it was also used for pedestrian detection in movies and videos as well as the vehicle and common animal detection in static images. However, the impact on performance of human face identification by parameters of
HOG has not been with detailed analysis [4]. Therefore, based on the extraction and identification of HOG-based raised or indented characters, this paper attempts to discuss and analyze the influence of each parameter.
