Edge width and edge gradient: influence on accuracy when measuring areas in lossily compressed images

Abstract In this paper, we evaluate the effect of lossy compression on luminance-based area measurements in images. On a set of 1,800 images, we vary three parameters: compression quality, edge width, and edge gradient, of the luminance edge bounding the area. As a result, when using the JPEG codec, the root-mean-square error after compression approximately doubles when decreasing the edge width from 16 to 4 pixel while keeping the gradient fixed. Conversely, for a fix width, the root-mean-square error doubles when halving the gradient. We also compared JPEG with the H.264 codec using default settings. H.264, based on inter-frame compression, requires four to six times the bitrate to achieve the results of JPEG. We conclude that for area measurements like these, among lossy codecs, intra-frame compression like JPEG is the best choice. Whether the influence of the compression is relevant depends on individual requirements, and more on the edge gradient than the edge width. Generally, we found that for steep gradients and reasonable JPEG compressions of Q = 85 or higher, our results were acceptable.