Combining Color and Wavelet Transform based Statistical Signatures Towards Content Based Image Retrieval
Keywords:Color Moment; Co-occurrence Signatures; Discrete Wavelet Transform; Euclidean distance; Kurtosis; Skewness.
This paper describes a Content Based Image Retrieval technique based on combined color and texture
feature. The color features are extracted in the form of weighted color moments using Hue, Saturation and Value (HSV)
color space. The texture information is characterized by exploring the statistical properties of wavelet detail coefficients.
Three different types of signatures are extracted: Energy signatures to estimate the distribution of energy at different
frequencies and scales, Histogram signatures to describe the shape of histogram for detail coefficients, and Cooccurrence signatures as a measure of gray level distribution. Euclidean distance is used to search relevant images
based on each of the above measures. Finally an integrated similarity index is computed to refine the results and retrieve
top 30 images relevant to the user’s query. The experimental results show the effectiveness of the proposed technique
over existing techniques using various other measures to describe texture and shape. The propose d method gives an
average accuracy of 71 % on Wang’s Image Database.