We perform a comparative evaluation of four feature descriptors for the task of feature matching in panorama stitching of images taken from architectural scenes and archaeological sites. Such scenes are generally characterized by structured objects that vary in their depth and large homogeneous regions. We test SIFT, LIOP, HRI and HRI-CSLTP descriptors on four different categories of images: well structured with some depth variations, partially homogeneous with large depth variations, nearly homogeneous with a little amount of structural details and illumination-variant. HRI-CSLTP and SIFT perform on par with each other and are better than the others on many of the test scenarios while LIOP performs well when the intensity changes are complex. We also study techniques used in measuring the dissimilarity of descriptors. Cross-bin measures that compare the value of a bin with all the other bins are useful when the changes induced on the descriptors are complex. Nevertheless, ‘heavy’ bins which represent common but insignificant features can dominate the computation of the dissimilarity values while ‘light’ bins that represent rare, yet significant features can be suppressed. We propose a cross-bin comparison technique that measures the dissimilarity between the given descriptors by boosting the significance of ‘light’ bins and suppressing the significance of ‘heavy’ bins. The proposed technique is tested on the problem of SIFT keypoint matching when changes in viewpoint, illumination and blurring occur and is found to perform better than the baseline dissimilarity measures like the Earth mover’s, Euclidean and the Chi-squared distances. © Springer Nature Singapore Pte Ltd. 2017.