Research
A brief introduction to the projects I am working on.
Image Analysis of food microstructure
Quality control in food research is often carried out by careful evaluation of microstructural features such as holes, protein, and fat of food samples. Indeed, the microstructural arrangement of basic elements of a complex food product affects its macroscopic behaviour and properties.
Microscopy combined with quantitative analysis is often used to find out the relationships between food properties and its microstructure. However, segmentation of structural features in food images is mostly carried out with time-consuming user interaction, by means of techniques as simple as thresholding and basic morphological operators.
Reliable quantitative analysis of microscopic features can be automatically accomplished only if a high-quality segmentation is computed beforehand. The measurements computed on binary images are deeply influenced by the segmentation accuracy, especially when shape descriptors of features are involved.
This project aims at developing high-quality segmentation algorithms for food microscopy applications, as well as advanced tools to analyse image data.
Medical Image Segmentation and Analysis
The interaction between medicine and computer science has brought notable improvements to the execution of common clinical and surgical problems. Computer-assisted medical procedures allow to monitor pathologies and, notably, to evaluate the outcome of medical cares.
In this scenery, digital imaging empowers physicians and scientists with new insights on the internal and external structures of human body. Ad-hoc software tools can be of great help to biologist, surgeons, and physicians for scientific observation and manipulation of patients’ data.
The main goal of medical image analysis is finding solutions to to problems related to the extrapolation, processing, visualization, and retrieval of new and non-trivial knowledge from medical imagery, as well as to provide new paradigms to approach known problems.
This area of investigation involves multidisciplinary efforts from computer scientists, physicians and industrial partners, to design, assess, and experiment with new techniques and to devise effective solutions to clinical problems.
High Dynamic Range and Tone Mapping of images and video
The human visual system can handle dynamic ranges that are several orders of magnitude larger than those of conventional acquisition and visualization devices. In order to fill the gap between the direct observation of a scene and its digital representation, high dynamic range sensors have recently been devised. Techniques are also used to build high-dinamics images from the output of low dinamics devices, e.g. by means of multi-esposure.
Recently, a number of systems have been proposed to extend the dynamic range of current visualization devices. However, the problem is far from being well investigated. As a consequence, the dynamic range of high-dynamics images must be reduced to fit the one of the visualization device at hand.
Several solutions exist for the tone mapping problem. However, most of them only cope with still images. These algorithms cannot be simply extended to video. Moreover, in driving assistance applications, video processing is usually performed on low-cost hardware, with devices often embedded in the camera box itself (e.g., smart cameras).
This project addresses the problem of tone mapping of high-dynamics video sequences. Special attention is put to temporal illumination variations.
Automatic 3D model acquisition using surface scanners
The models produced by means of the available 3D-surface scanning technologies are considered accurate enough for most applications. Unfortunately, the acquisition of complex objects is still a demanding and human-intensive process that cannot be performed by non-specialists.
Among the open problems, one of the most difficult to grasp is planning the acquisition session, i.e. choosing a set of scanner positions to view the whole surface of the object.
Another human-intensive phase is surface registration, and a painful one if large objects are being imaged.
This research addresses both issues of registration of range maps and sensor planning, in the context of 3D model acquisition.
Image compression
Image compression is one of the most common and studied problems in the image processing domain, and an ubiquitous one.
Although a huge amount of work has been made in this field, there are still several open problems.
I studied two unusual aspects in image compression: indexed images, and pre-processing optimisation.