Automatic South African Coin Recognition Through Visual Template Matching
Tharish Sooruth, Mandlenkosi Victor Gwetu · 2018
Since money is an integral part of our everyday lives and it has an attached economical value, it is essential for it to be counted accurately. Although previous research based on a visual approach towards automatic coin counting has been proposed in order to circumvent human error and facilitate records for auditing, there is still room for improvement and further application. This study analyzes the effect of edge and texture features on visual template matching in the context of South African coin recognition. Image processing techniques are used to extract three features: radius, color and texture; from coin images for use in automatic identification. These features have been identified as the possible main differentiators between the existing South African coins. The radius is computed through the Circular Hough Transform (CHT), average color is calculated through K-Means clustering, and texture is derived from spatial histograms on Local Binary Patterns (LBP). Rotation invariance is achieved by transforming a coin image from Cartesian to Polar coordinate form. Coin template matching is done using the Chi-Square distance measure and Normalized Cross Correlation (NCC). Experiments on a sample South African coin dataset yielded a recognition rate of 91.67%.