Image hashing by LoG-QCSLBP

This paper presents an image hashing algorithm for authentication and tampering based on texture features. Center Symmetric Local Binary Pattern (CSLBP) feature is computationally simple, rotation invariant which works in spatial domain. In CSLBP, number of histogram bin for each sub block of an image is 16, unlike 256 bin in Local Binary Pattern (LBP). In our proposed method, flipped difference is used to generate a histogram of only 8 bin, for each sub block. Resultant method with 8 bin histogram has less discrimination power. To enhance discrimination power, Laplacian of Gaussian (LoG) is used as a weight factor during histogram construction. LoG is used to find a characteristic scale for a given image location. LoG is a second order derivative edge detection operator which performs well in presence of noise. In our previous papers, we tried various local descriptors like magnitude of difference, standard deviation, coefficient correlation as a weight factor, to enhance the success rate of compressed CSLBP. Proposed LoG-QCSLBP gives good results for JPEG, salt & pepper noise, brightness plus, increase/decrease contrast. In the results section, we compared all variants of compressed CSLBP. Results clearly show that by incorporating the weight of a local descriptor, discrimination power of compressed CSLBP is enhanced.