Prof. Dr. Hab. Jean-Charles PINOLI |
|
All publications of J.-C. Pinoli:
An image segmentation process often results in a special spatial set, called a mosaic, as the subdivision of a domain S within the n -dimensional Euclidean space. In this paper, S will be a compact domain and the study will be focused on finite Jordan mosaics, that is to say mosaics with a finite number of regions and where the boundary of each region is a Jordan hypersurface. The first part of this paper addresses the problem of comparing a Jordan mosaic to a given reference Jordan mosaic and introduces the ε dissimilarity criterion. The second part will show that the ε dissimilarity criterion can be used to perform the evaluation of image segmentation processes. It will be compared to classical criterions in regard to several geometric transformations. The pros and cons of these criterions are presented and discussed, showing that the ε dissimilarity criterion outperforms the other ones. |
Summary The logarithmic image processing (LIP) model is a mathematical framework which provides algebraic and functional operations for the processing of intensity images valued in a bounded range. The LIP model has been proved to be physically consistent, most notably with some image formation models and several laws and characteristics of human brightness perception. This paper addresses the image focus measurement problem using the LIP model. The three most classical image focus measurements: the sum-modified-Laplacian, the tenengrad and the variance, which aim at estimating the degree of focus of an acquired image by emphasizing and quantifying its sharpness information, are considered and reinterpreted using the LIP framework. These reinterpretations notably make attempts at evaluating degrees of focus in terms of human brightness (sensation) from physical light stimuli. Their potential is illustrated and validated on shape-from-focus issues on both simulated data and real acquisitions in digital optical microscopy. The concept of shape-from-focus involves recovering the shape of an observed thick sample by locally maximizing a focus measurement throughout a sequence of differently focused images. Finally, it is shown that the LIP-based focus measurements clearly outperform their respective classical ones. |
Purpose Corneal endothelium of organ cultured cornea forms a 3D surface because of corneal curvature and of the Descemetic folds (of hundreds of μm depth) that appear during storage. Eye bank technicians have therefore a risk of endothelial cell density (ECD) overestimation when counting on an image projection. Our aim is to develop a 3D reconstruction from a Z stack of images in order to take account of the local slope and increase ECD reliabilityMethods Z stacks of 35 images were acquired with a motorized light microscope (BX41, Olympus) through a x10 objective. After routine eye bank preparation with saline, EC appeared with a clear cytoplasm and dark borders but could locally show a contrast reversal. In each section, the best-focused pixels were determined using focus measurement algorithms. Depth map and texture image were then separately generated and the 3D reconstruction was obtained by projecting the texture image onto the depth map. A new algorithm, based on projections onto eigenbases, was developed and compared to 5 existing algorithms. Comparison was done using 1/ simulated images with additive noises to assess the reliability and noise robustness 2/ true endothelial imageResults The new algorithm exhibited a more robust behaviour in presence of noise than the other algorithms. 3D endothelial reconstructions were also visually better, especially in case of local contrast reversalConclusion Using a technology easily transposable into an eye bank, a reliable 3D reconstruction is possible from routine images that are often of average quality with poor contrast and high background noise. Our next step is to integrate the local slope information to correct local ECD |
In human retina observation (with non mydriatic optical microscopes), a registration process is often employed to enlarge the field of view. For the ophthalmologist, this is a way to spare time browsing all the images. A lot of techniques have been proposed to perform this registration process, and indeed, its good evaluation is a question that can be raised. This article presents the use of the epsilon dissimilarity criterion to evaluate and compare some classical featurebased image registration techniques. The problem of retina images registration is employed as an example, but it could also be used in other applications. The images are first segmented and these segmentations are registered. The good quality of this registration is evaluated with the epsilon dissimilarity criterion for 25 pairs of images with a manual selection of control points. This study can be useful in order to choose the type of registration method and to evaluate the results of a new one. |
{A} method for spatial registering pairs of digital images of the retina is presented, using intrinsic feature points (landmarks) and dense local transformation. {F}irst, landmarks, i.e. blood vessel bifurcations, are extracted from both retinal images using filtering followed by thinning and branch point analysis. {C}orrespondances are found by topological and structural comparisons between both retinal networks. {F}rom this set of matching points, a displacement field is computed and, finally, one of the two images is transformed. {D}ue to complex retinal registration problem, the presented transformation is dense, local and adaptive. {E}xperimental results established the effectiveness and the interest of the dense registration method. |
The cell density in the human corneal endothelium is mainly responsible for the transparency of the cornea. Its measurement is thus of importance for the ophthalmologists, that classicaly use optical specular microscopy to capture in vivo images of the corneal endothelium. An image segmentation process is then used to detect the cells and partition the images. Different computational methods can perform this task, but it remains visually difficult to evaluate the quality of their results. This article proposes a novel quantitative evaluation criterion, adapted to the cellular spatial structure (called a mosaic, that results of the segmentation). This criterion does not respect the properties of a metric, but is related to the notion of dissimilarity defined by the psychologists. It is used to quantify the results of three dedicated image segmentation methods on a learning image database and thus allows a first classification between them. The two best methods are hence selected and then applied on hundred new images. They are blindly proposed to the ophthalmology experts for marking. Their judgement corroborates the quantitative analysis, which confirms the practical relevance of this dissimilarity criterion. |
The human visual system is far more efficient than a computer to analyze images, especially when noise or poor acquisition process make the analysis impossible by lack of information. To mimic the human visual system, we develop algorithms based on the gestalt theory principles: proximity and good continuation. We also introduce the notion of mosaic that we reconstruct with those principles. Mosaics can be defined as geometry figures (squares, triangles), or issued from a contour detection system or a skeletonization process. The application presented here is the detection of cornea endothelial cells. They present a very geometric structure that give enough information for a non expert to be able to perform the same analysis as the ophthalmologist, that mainly consists on counting the cells and evaluating the cell density. |
In human retina observation (with non mydriatic optical microscopes), a registration process is often employed to enlarge the field of view. For the ophthalmologist, this is a way to spare time browsing all the images. A lot of techniques have been proposed to perform this registration process, and indeed, its good evaluation is a question that can be raised. This article presents the use of the epsilon dissimilarity criterion to evaluate and compare some classical featurebased image registration techniques. The problem of retina images registration is employed as an example, but it could also be used in other applications. The images are first segmented and these segmentations are registered. The good quality of this registration is evaluated with the epsilon dissimilarity criterion for 25 pairs of images with a manual selection of control points. This study can be useful in order to choose the type of registration method and to evaluate the results of a new one. |