In some cases, consumer devices (monitors, printers, cameras, etc) do not allow us to capture or reproduce the full range of luminance information of a synthetic or real scene. Many techniques have been proposed in the last years, that address the tone mapping problem (i.e. the conversion from real luminance values into available ranges) trying to simulate some mechanisms of the Human Visual System, but few of these take into account the color constancy problem, i.e. the ability to discount color cast induced by illuminant. In this paper we present some experiments on high dynamic range images, modifying a Retinex implementation, the Brownian Retinex algorithm (characterized by the construction of Brownian random paths in the image), to solve the tone mapping and the color constancy problem at the same time. We also try to better simulate the process of scene observation, considering the eye movements called saccades and simulating them by applying different statistical distributions to random paths construction. Moreover, we propose an alternative technique to compute Retinex ratios on the paths, based on the assumption that vision can be considered a kind of sampling process.