Comprehensive Performance Comparison of Cosine, Walsh, Haar, Kekre, Sine, Slant and Hartley Transforms for CBIR With Fractional Coefficients of Transformed Image

The desire of better and faster retrieval techniques has always fuelled to the research incontent based image retrieval (CBIR). The extended comparison of innovative content basedimage retrieval (CBIR) techniques based on feature vectors as fractional coefficients oftransformed images using various orthogonal transforms is presented in the paper. Here thefairly large numbers of popular transforms are considered along with newly introducedtransform. The used transforms are Discrete Cosine, Walsh, Haar, Kekre, Discrete Sine,Slant and Discrete Hartley transforms. The benefit of energy compaction of transforms inhigher coefficients is taken to reduce the feature vector size per image by taking fractionalcoefficients of transformed image. Smaller feature vector size results in less time forcomparison of feature vectors resulting in faster retrieval of images. The feature vectors areextracted in fourteen different ways from the transformed image, with the first being all thecoefficients of transformed image considered and then fourteen reduced coefficients sets areconsidered as feature vectors (as 50%, 25%, 12.5%, 6.25%, 3.125%, 1.5625% ,0.7813%,0.39%, 0.195%, 0.097%, 0.048%, 0.024%, 0.012% and 0.06% of complete transformedimage coefficients). To extract Gray and RGB feature sets the seven image transforms areapplied on gray image equivalents and the color components of images. Then these fourteenreduced coefficients sets for gray as well as RGB feature vectors are used instead of using allcoefficients of transformed images as feature vector for image retrieval, resulting into betterperformance and lower computations. The Wang image database of 1000 images spreadacross 11 categories is used to test the performance of proposed CBIR techniques. 55queries (5 per category) are fired on the database o find net average precision and recallvalues for all feature sets per transform for each proposed CBIR technique. The results haveshown performance improvement (higher precision and recall values) with fractionalcoefficients compared to complete transform of image at reduced computations resulting infaster retrieval. Finally Kekre transform surpasses all other discussed transforms inperformance with highest precision and recall values for fractional coefficients (6.25% and3.125% of all coefficients) and computation are lowered by 94.08% as compared to Cosine orSine or Hartlay transforms.