Still images can be acquired by taking multiple exposures of the same scene and combining them together; video sequences can be acquired by means of appropriate HDR cameras, typically based on CMOS sensors. Appropriate file formats have been developed that are able to encode the actual photometric values of an HDR image, typically using a floating-point representation. In order to visualize an HDR image on a conventional display or on print, it is possible to use appropriate dynamic range reduction techniques. One remarkable property of the HVS is that the sensation of brightness is not determined by the physical luminance of the fixated point, but rather on the spatial or temporal luminance variation; as a consequence, a processing that adjusts the large-scale luminance without altering the fine details can reduce the dynamic range of an image – thus allowing its reproduction on a conventional display – without altering its appearance and information content. The following figure shows the same scene of the previous example, mapped with an advanced dynamic range reduction algorithm developed at the Image Processing Laboratory: the details in all the portions of the image are visible and the image maintains a natural appearance; in most cases, an observer is unaware that the image has been processed at all.
Dynamic range reduction algorithms are an active research topic in the academic and scientific community; in recent times, the developed techniques are beginning to appear also in consumer software applications, and examples of HDR photography are often posted on photography websites and forums. Simple techniques can introduce different kinds of artifacts, such as halos around the sharp edges, over-enhancement of the details, noise or alterations of the colors. The design of a high-quality algorithm requires advanced knowledge in different topics such as photometry and color science, the physiological and psychological aspects of vision, calculus, numerical analysis and computer science. The quality of the processed image can be assessed and tuned either subjectively – by means of psychophysical experiments in which the users give a rating to the processed images or adjust some parameters in the algorithm – or objectively – by means of analytical expressions that measure specific kinds of artifacts. Besides the visual quality of the processed images, desirable features of a dynamic range reduction algorithm are the computational efficiency and the simple tuning of the parameters.
The processing of video sequences poses additional issues in two main aspects. In most cases, it is not possible to simply process each frame individually, as this would likely introduce a visible and annoying flicker in the processed video. It is therefore necessary to add a dedicated inter-frame processing that guarantees the temporal coherence. Furthermore, if a real-time processing is requested, it is necessary to simplify the algorithms and to implement them on dedicated hardware, such as field-programmable gate arrays (FPGAs), capable of providing the necessary processing power.
Contact: prof. G. Ramponi (