Skip to main content
European Commission logo print header

Advanced modelling of visual information processing

Deliverables

The outlined mechanism can be classified as a cellular automation. Essentially, it consists of a set of differential equations. The equations are exclusively coupled by means of next-neighbour-interactions. The process can be implemented in VLSI, and utilised for a wide range of circuits in which there is a need for normalisation of an input signal for a successive processing stage, which has a confined dynamic range. In its simplest form, the method consists of three differential equations. Each equation can be understood as an independent layer with an ordered array of cells. The topology is the same for all three layers. The first layer thereby performs a nonlinear diffusion and converges with each of its cells to the global maximum. This means that the global maximum is locally available, despite of mere local interactions of the cells in the layer. Analogously, the second layer converges locally to the global minimum. The third layer connects both the minimum and the maximum layer in order to produce a normalised representation of the input. The method proposed provides a definite solution to the problem of searching the global maximum and the global minimum of a set of numbers by only using local and non-algorithmic interactions. "Non-algorithmic" means that we do not use explicit comparisons between values (e.g. if...then). "Local" means that we have no globally available memory or buffer, but only allow interactions between neighbouring cells. About potential applications the network can be used i.e. for image fusion for tasks like night vision, target recognition, security, law enforcement, etc. Potential end-users of this result are consumer electronics, autonomous navigation, surveillance, etc. The main benefits of the current result against 'classical' approaches is that the later ones demand sophisticated mechanisms to process and map images of different spatial resolution and with different dynamics/ranges in their discretised values. Source code was written in Matlab and C++.
A method to evaluate the performance of a wide range of video transmission equipment, such as recorders, format converters, mixers, etc. has been developed. The method evaluates both subjective and objective (numerical) image quality. A system based on this method is currently used by TELEVISA, probably the largest television broadcasting enterprise in Latin America. This system allows the company to choose among the equipment that different manufacturers offer.
A novel scheme for image compression was developed. This scheme offers compression rates similar to JPEG, but with improved perceptive quality. At current time, experiments are being performed to transmit video images over Internet 2. A number of potential applications include videoconference and in general all applications that need to send visual information over Internet.
A system for blur estimation has been developed. It works stand-alone, or can be incorporated into an existing image processing toolbox. At the present time, a specific application for the Mexican Federal Electoral Institute is under development. This institute is interested in an automatic system that allows them to detect the photos that were taken defocused and that are meant to appear in the elector identity card that every Mexican citizen needs in order to vote. Yearly, this institute wastes a lot of money in identity cards that are lately rejected because of poor photo quality. An automatic defocus estimator would prevent the card from being printed in the case of blur in the elector photo.

Exploitable results

The Hermite Transform has proved to be an efficient image representation model to build different algorithms for image processing. In the area of Remote Perception applications include noise (speckle) reduction, segmentation, classification and fusion. At present time, this tool is being used by the Center for Geography and Geomatics Research, CONACYT, Mexico. This Center focuses its activities in projects for government and industry in the field of Remote Perception, Cartography, Geographic Information Systems, etc. i.e. in all areas that include Geography and Informatics. During a year, starting in September 2002, cooperation will be launched in order to develop a series of tools for processing Synthetic Aperture Radar and LANDSAT images based on the Hermite Transform. The applications will include noise reduction, segmentation, classification and fusion. The latter will be included within toolboxes of software packages that are commonly used in the Center, such as PCI, ERDAS, ENVI, MATLAB, etc. A first project in which this tools will be used is the classification of vegetation areas in Mexico City, as requested by the government of this city.
The Hermite Transform is an image representation model that incorporates some important properties of visual perception such as the analysis through overlapping receptive fields and the Gaussian derivative model of early vision. It also allows the construction of pyramidal mutirresolution analysis-synthesis schemes. Based on this transform, we have built image fusion schemes that take advantage of the fact that Gaussian derivatives are good operators for the detection of relevant image patterns at different spatial scales. These patterns are later combined in the transform coefficient domain. Applications of this fusion algorithm have been tested on remote sensing images, namely LANDSAT, IKONOS, RADARSAT and SAR AeS-1 images.

Searching for OpenAIRE data...

There was an error trying to search data from OpenAIRE

No results available