EFFECTIVENESS OF NON-PARAMETRIC TECHNIQUES IN IMAGE RETRIEVAL

The need to search for a particular image(s) of interest from an image database or unstructured image collection has led to the search of effective image retrieval systems. To achieve this, there is need to find segmentation, representation and similarity matching algorithms that work in harmony. This study looked at the non-parametric techniques in image retrieval system. The effectiveness of Epanechnikov, Gaussian and Histogram non-parametric algorithms were compared in a generic image retrieval system. The Chan & Vese and Cosine Angle Distance algorithms were used for segmentation and similarity matching respectively. The performance of the non-parametric techniques was measured using recall-precision curve and the Bull's Eye performance score. The estimating techniques performed better than the absolute value technique. The study then looked at the limitations of these techniques.