Finding Image Features Using Deformable Templates and Detailed Prior Statistical Knowledge

Much work in image processing has been devoted to generating filters to detect low level image features, e.g. edges, peaks, valleys. Objects are then located or recognised in the image by using the output from these filters. We present methods which can interpret the output from filters in order to locate known objects in images. The algorithms use statistical knowledge about the variation of the shape of objects being searched for in order to guide the search to only feasible regions. The main techniques used are various deformable template algorithms where optimisations are achieved by random sampling and simulated annealing to avoid non-global extrema. The particular application here is for locating facial features including head outlines, where the results give key locations on the face and allow approximate geometric representations of the features to be reconstructed.