PERFORMANCE OF PHASE-BASED ALGORITHMS FOR DISPARITY ESTIMATION

Stereoscopic depth analysis by means of disparity estimation has been a classical topic of computer vision, from the biological models of st ereopsis [1] to the widely used techniques based on correlation or sum of squared differences [2]. Most of the recent work on this topic has been devoted to the phase-based techniques, developed because of thei r superior performance and better theoretical grounding [3, 4]. In thi s article we characterize the performance of phase-based disparity est imators, giving quantitative measures of their precision and their lim its, and how changes in contrast, imbalance, and noise in the two ster eo images modify the attainable accuracy. We find that the theoretical range of measurable disparities, one period of the modulation of the filter, is not attainable: the actual range is approx. two-thirds of t his value. We show that the phase-based disparity estimators are robus t to changes in contrast of 100% or more and well tolerate imbalances of luminosity of 400% between the images composing the stereo pair. Cl earing the Gabor filter of its DC component has been often advocated a s a means to improve the accuracy of the results. We give a quantitati ve measure of this improvement and show that using a DC-free Gabor fil ter leads to disparity estimators nearly insensitive to contrast and i mbalance. Our tests show that the most critical source of error is noi se: the error increases linearly with the increase in noise level. We conclude by studying the influence of the spectra and the luminosity o f the input images on the error surface, for both artificial and natur al images, showing that the spectral structure of the images has littl e influence on the results, changing only the form of the error surfac e near the limits of the detectable disparity range. In conclusion, th is study allows estimation of the expected accuracy of custom-designed phase-based stereo analyzers for a combination of the most common err or sources.