Giuliana Ramella

Finding contour-based abstractions of planar patterns

Abstract An algorithm is described to detect a number of points, on the contour of a planar shape, which constitute the vertices of a schematic polygonal representation of the shape itself. A set of points, initially extracted from the chain-coded representation of the contour, is iteratively examined, while removing some points and inserting new ones. The number of selected points decreases in size from iteration to iteration, and the selection process converges towards an expected perceptually significant set of points. The polygon obtained by linking successive points approximates the contour in an intuitive way. It is not constrained within a given tolerance, and is likely to locally change from a coarse to a more faithful approximating shape, in correspondence with contour regions increasing in details.

Finding grey-skeletons by iterated pixel removal

Abstract The grey-skeleton is understood as a connected subset of a grey-scale pattern, which is a stylized version consisting of a network of digital lines centrally placed along local higher intensity regions. We present a parallel thinning algorithm that relies on the iterated erosion of the pattern, and which proceeds from lower grey values towards higher ones until the grey-skeleton is finally obtained. The process includes a preliminary phase in which the significance of the hollows and plateaux possibly existing in the pattern is investigated. In particular, the hollows with a significant depth are regarded as topological constraints for the skeleton structure.

Sketching a grey-tone pattern from its distance transform

We describe a procedure to create an abstraction of a grey-tone pattern by sketching its regions which have locally higher intensities. The sketch is a set of simple digital lines, qualitatively analogous to the skeleton representation computed in the case of a single-valued pattern. The grey-tone pattern is regarded as constituted by a number of regions with constant grey-value, and the skeleton is found by detecting suitable pixels on the Distance Transform of the pattern. Computation of the Distance Transform is accomplished according to the city-block distance, by ordered propagation over regions with increasing grey-values. Neighbourhood conditions are used to detect the set of the skeletal pixels, which is subsequently reduced to unit thickness. Finally, the skeleton undergoes a pruning process which removes a part or all of some of its branches.

Shape and topology preserving multi-valued image pyramids for multi-resolution skeletonization

Starting from a binary digital image, a multi-valued pyramid is built and suitably treated, so that shape and topology properties of the pattern are preserved satisfactorily at all resolution level ...

A fully automatic one-scan adaptive zooming algorithm for color images

We present an interpolation algorithm for adaptive color image zooming. The algorithm produces the magnified image in one scan of the input image, and is fully automatic since does not involve any a priori fixed threshold. Given any integer zooming factor n, each pixel of the input image generates an nxn block of pixels in the zoomed image. For the currently visited pixel of the input image, the pixels of its associated block are first assigned tentative values, which are then adaptively updated before building the next block. The method is suggested for RGB images, but can equally be employed in other color spaces. Peak signal to noise ratio (PSNR) and Structural SIMilarity (SSIM) are used to evaluate the performance of the algorithm.

Oversegmentation reduction via multiresolution image representation

We introduce a method to reduce oversegmentation in watershed partitioned images, that is based on the use of a multiresolution representation of the input image. The underlying idea is that the most significant components perceived in the highest resolution image will remain identifiable also at lower resolution. Thus, starting from the image at the highest resolution, we first obtain a multiresolution representation by building a resolution pyramid. Then, we identify the seeds for watershed segmentation on the lower resolution pyramid levels and suitably use them to identify the significant seeds in the highest resolution image. This is finally partitioned by watershed segmentation, providing a satisfactory result. Since different lower resolution levels can be used to identify the seeds, we obtain alternative segmentations of the highest resolution image, so that the user can select the preferred level of detail.

Grey Level Image Components for Multi-scale Representation

A method to identify grey level image components, suitable for multi-scale analysis, is presented. Generally, a single threshold is not sufficient to separate components, perceived as individual entities. Our process is based on iterated identification and removal of pixels, with different grey level values, causing merging of grey level components at the highest resolution level. A growing process is also performed to restore pixels far from the fusion area, so as to preserve as much as possible shape and size of the components. In this way, grey level components can be kept as separated also when lower resolution representations are built, by means of a decimation process. Moreover, the information contents of the image, in terms of shape and relative size of the components, is preserved through lower resolution representations, compatibly with the resolution.

Color Quantization by Multiresolution Analysis

A color quantization method is presented, which is based on the analysis of the histogram at different resolutions computed on a Gaussian pyramid of the input image. Criteria based on persistence and dominance of peaks and pits of the histograms are introduced to detect the modes in the histogram of the input image and to define the reduced colormap. Important features of the method are, besides its limited computational cost, the possibility to obtain quantized images with a variable number of colors, depending on the users need, and that the number of colors in the resulting image does not need to be a priori fixed.

Using Top-Down and Bottom-Up Analysis for a Multiscale Skeleton Hierarchy

Multi-scale skeletons can be conveniently employed in the matching phase of a recognition task. The multi-scale skeletons are here obtained by first computing the skeleton at all levels of a resolution structure and then establishing a hierarchy among skeleton components at different scales, using a parent-child relationship. Although subsets of the skeleton expected to represent given pattern subsets may consist of different number of components at different scales, a component preserving decomposition is obtained that produces a hierarchy in accordance with human intuition.

Color histogram-based image segmentation

An algorithm is presented to segment a color image based on the 3D histogram of colors. The peaks in the histogram, i.e., the connected components of colors with locally maximal occurrence, are detected. Each peak is associated a representative color, which is the color of the centroid of the peak. Peaks are processed in decreasing occurrence order, starting from the peak with the maximal occurrence, with the purpose of maintaining only the representative colors corresponding to the dominant peaks. To this aim, each analyzed peak groups under its representative color those colors, present in the histogram and that have not been grouped to any already analyzed peak, such that their distance from the centroid of the peak is smaller than a priori fixed value. At the end of the grouping process, a number of representative colors, generally substantially smaller than the number of initial peaks, is obtained, which are used to identify the regions into which the color image is segmented. Since the histogram does not take into account spatial information, the image is likely to result over-segmented and a merging step, based on the size of the segmentation regions, is performed to reduce this drawback.