Autocorrelation based Chordiogram Image Descriptor for Image Retrieval
A. Saravanan, Sarmitha Sathiamoorthy · 2019
This paper proposes an autocorrelation based chordiogram image descriptor (ACID) for image searching and retrieval. The standard 4D chordiogram image descriptor is a ordered collection of local edgel chordiogram (LEC) of all the patches of image and it encodes length of the line segment among the pair of edgels, orientation of each edgel and degree of angle among the line segment and horizontal plane for every pair of predominant edgels in a patch whereas proposed approach uses autocorrelation for computing the LEC and it exploits the spatial correlation and above mentioned features of standard LEC among the identical pair of edgels at distance d. The novelty of proposed CID is, it captures localized texture, spatial details besides to the geometric details offered by the standard CID except length of the line segment which is 1 in the proposed approach. The proposed ACID leads to better retrieval with resilient to illumination, scaling, translation, noise and small rotation. In our proposed approach, geometric details also more localized one. The experiments have been done for demonstrating the worth of proposed ACID on four different types datasets: (i). Gardens Point Walking, (ii) St. Lucia, (iii) UA Campus, (iv) Self photographed image. The results confirm the supremacy of proposed ACID approach by means of average retrieval precision and average retrieval rate over the state-of-the-art approaches.