Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
In this paper we propose a low complexity method for Rotation, Scale and Translation (RST) invariant content-based image retrieval, suitable for a handheld image recognition device. The RST compensation method is based on Fourier-Mellin Transform (FMT) which we implement efficiently using log-polar grid interpolation. This RST compensation method is used in conjunction with an image recognition algorithm based on Discrete Cosine Transform (DCT) phase matching. A pre-selection algorithm is also added for decreasing the complexity. This algorithm is based on color proportions within concentric circular zones encompassing the edge pixels. The resulting RST invariant image recognition system was tested on 1500 pictograms and 1000 pictures with different RST conditions, showing an average recognition accuracy of 95.2% for pictograms and 96.9% for pictures.
Fabio Nobile, Giovanni Migliorati