Summary: The objective of image fusion is to represent relevant information from multiple individual images in a single image. Some fusion methods may represent important visual information more distinctively than others,thereby conveying it more efficiently to the human observer. Hence, there is a growing need for metrics to evaluate and compare the visual quality of fused imagery. Here we propose to rank order images fused by different methods according to the computational attention value of their visually important details. First we compute for each of the fused images a multi-bitrate attention map, following a rational model of computational attention. From this attention map, we then calculate the average attention score within areas of interest (e.g., living creatures, man-made objects, and terrain features), for each bitrate. A high computed mean attention value within the areas of interest at any reconstruction fidelity corresponds to a high computational saliency of the areas of interest. If the computational results agree with human observer performance, it makes the approach valuable for practical applications. We shall introduce a method to rank order images according to the computational attention value of their visually important details. We will apply the method to rank different sets of fused and input images in the order of important information visibility. Visual efficiency will be predicted by means of the normalized mean attention within regions of interest either provided by human observers or by the use of automated detection. A multi-bitrate attention map will provide us with a computational attention score for each spatial location at high and low quality versions of the image reconstruction. The main advantages of this approach for comparative visual efficiency analysis will be its simplicity and speed. Here we will develop an implementation of the rational model of computational attention whose interest will be not only in cost reduction but also in a real-time analysis of new images. We are going to develop a publicly available suite of Web-based comparative visual efficiency tools designed to facilitate comparison of fused images. In addition we will provide an interface for comparative visual efficiency analysis, which like all of the tools reported here, will be freely available to the scientific community.

Summary: In any efficient transmission system there must be a number of bitstreams with low-cost transmission that enter into a prioritization protocol and which are used up. The available low-cost bit streams cannot be increased. In the long run the limited availability of these exhaustible (low-cost) bit streams would begin to act as a constraint on the performance of the transmission system. Although this is an obvious point most studies of the properties of the optimal transmission plans neglect it. In this project, we will analyze the problems that appear to rise when the depletion of low-cost bitstreams is incorporated into the study of intertemporal transmission plans.To this aim we will develop an analytical discussion of the availability of low-cost bits in progressive transmission. The models of bit depletion are to be formulated so that it can be solved by methods from control theory. The analytical results could display key aspects of the use of exhaustible (low-cost) bit resources on the optimal transmission path, for example: (i) As the weight given to future profits increases we keep more low-cost bit resources for the future bit allocation; (ii) the easier it is to substitute knowledge about relevant visual information for low-cost bit resources and the more important is knowledge for prioritization, the more we substitute knowledge for low-cost bit resources; and (iii) if the rent component of profit is the difference between the profit from bit allocation and the average transmission cost, under the assumption that bit transmission costs are bounded above, this rent becomes zero as the data of low-cost transmission is exhausted. This project will study which are the practical applications of these (or others) analytical results on the optimal transmission path and how they can be used to work on real transmission systems.

Summary: The effectiveness of a coding method can be improved through a space-varying filterbank tree representation of the image, and this property can be conveniently exploited using appropriate bit allocation strategies among the spatial segments of the image. Here we examine the conditions for achieving a rational agreement on the distribution problem by stating axioms that its solution must obey in absence of a priori knowledge about regions of interest. Firstly, a measure of benefit avoiding certain forms of behavioral inconsistency is to be assigned to each possible bit allocation in such a way that each region's preference may be inferred between any two bit allocations from their respective benefits. Secondly, individual regions are to agree on an allocation of bits which is then to be brought about by a joint strategy, but, under what conditions is their agreement rational? Here we will propose a characterization of rational agreement whose solution is an application of a general procedure for cooperative action where each may benefit only on terms which permit proportionately equal benefits to others. Experimental results will be given to evaluate the performance of the strategy of collective rationality for the allocation of bits, based upon a validated predictor for visual distinctness from digital imagery. The general objective is to demonstrate that, in absence of a priori knowledge about regions of interest, this novel scheme for bit allocation can be used with reasonable internal consistency in order to achieve image fidelity superior to that of the state of art in image compression.

Summary: The RGFF representational model can be used to decompose the original scene into a number of images isolating statistical structures which maximize the redundancy across scales and orientations. The derived redundancies can then be exploted to decrease the number of bits that are needed to code the original scene. But the problem of lossy image coding techniques is that there is a trade-off between image distortion and coding rate. This trade-off may be reached with several techniques, but all of them require an ability to measure distortion. And finding a general enough measure of perceptual quality has proven to be an elusive goal. To circumvent the lack of knowledge of what distortion measures are more suitable for images, here we propose to develop a novel technique for deriving optimal performance bounds (it has been termed the "best achievable" compression ratio) based on the relationship between information theory and the problem of testing hypotheses. The best achievable compression ratio for lossy coders will determine a boundary between achievable and non-achievable regions in the trade-off between source fidelity and coding rate. The resultant bounds will be tight for situations of practical relevance (i.e., correspond to high coding rate). These performance bounds will be directly achievable by a constructive procedure as suggested in a theorem which is intended to prove the relationship between the "best achievable" compression ratio and the Kullback-Leibler information gain. We will test the best achievable compression ratio for various lossy coding schemes and several kind of scenes. We also evaluate the different coding techniques in the rate-distortion sense by using the concept of the best achievable compression ratio and the corresponding error probability in a bayesian setting. These results will provide an insight into the design issues of optimizing lossy coders, as well as a good reference for application developers to choose from an increasingly large family of lossy image coders for their applications

Summary: In this research project a novel approach to detection and distinction of objects in 2D digital images is developed. Instead of assuming that perceived objects are simple or statistical structure at a particular scale, we will regard them as visual patterns, distinguished at an object level. We wiil work with visual patterns as features which have the highest degree of congruence in statistical structure across different frequency bands. Under this frame we will analyze if it is possible to design methodologies to solve problems of target distinctness in complex natural images and feature extraction in biomedical images. Our definition of visual pattern will induce new representational models by using a minimal subset of activated recognizers sensitive to significant spatial structures in the scene. Its basic assumptions are to be in agreement with spatial-frequency channels models quite successful for the detection of visual patterns. The performance of the model is to be evaluated in terms of: (i) the increment of the signal-to-noise ratio in the resultant image features, (ii) the degree of decorrelation for perceptual features, (iii) the number of activated filters needed to analyze the visual scene, and (iv) the applicability of the representation for discrimination tasks on different kind of images

Summary: In this research project we postulate that the visual scene may be initially coded along a number of separable dimensions, such as color, orientation, spatial frequency, brightness, direction of movement, or stimulus locations. Each of these representations is registered early, automatically, and in parallel across the visual field. At a later stage, objects are identified separately as conjunctions of features. Conjunctions of features, on the other hand, are expected to require serial search and have no effect on performance unless focally attended; they should prove effective in the tasks of identification and location rather than in mediating texture segregation. The aim of this research project is to develop an artificial vision system which provides us with an organization of the image content according to conjuntions of features by recombining separable representations (color, form, brightness, etc) at stimulus locations with focal attention, in such a way that any features which are present in the same central "fixation" of attention are combined to form a single object. To deal with focal attention, a spatial sensitivity function should be incorporated that directs the construction of integral features to only significant locations of the reference picture. The resultant organization in conjunctions of features is used to create a general approach allowing the significant image forms to be successfully represented, analyzed and also compressed even if there is a lack of prior knowledge about the image content. Computational methods for the use of integral features for recognition, compression and image distortion quantification will be formulated, implemented and validated for a number of bio-medical images, images of natural scenes and face images

Summary: The aim of this research project is to develop an artificial vision system capable of dealing with biomedical digital images. Our approach to accomplish the task of visual perception as well as transmission and coding is motivated in part by recently discovered biological mechanisms of the human visual system. Multiresolution is a known feature of such system (there exist cortical neurons which respond specifically to stimuli within to certain frequencies). Several different formulations to deal with resolution in the image are here integrated to create a more general approach allowing the shapes to be successfully represented, analyzed and also compressed if there is a lack of prior knowledge about the image's resolution map. First, contrast enhancement of pure images as well as feature extraction that takes into account the level of detail corresponding to the image's intensity change models is considered. Second, a novel approach unifying all of the scale-based formulations in describing shapes is presented. Third, a new compression technique capable of removing the unwanted detail from noise, and simultaneously, of separately compressing each feature of interest at its proper scale is formulated. Such a compression technique also allows to selectively decode information at a specific resolution. In order to examine the effectiveness of the formulation proposed here, its performance has been evaluated for a number of bio-medical images. Hence, it can be deduced such a vision system determines an useful and efficient tool in assisting the biomedical expert