Leszek J. Chmielewski

Computer Vision and Graphics

Since most of even recently proposed image quality assessment metrics are typically applied for a single color channel in both compared images, a reliable color image quality assessment is still a challenging task for researchers. One of the major drawbacks limiting the progress in this field is the lack of image datasets containing the subjective scores for images contaminated by color specific distortions. After the publication of the TID2013 dataset, containing i.a. images with 6 types of color distortions, this situation has changed, however there is still a need of validation of some recently proposed grayscale metrics in view of their applicability for color specific distortions. In this paper some results obtained using different approaches to color to grayscale conversion for some well-known metrics as well as for recently proposed combined ones, are presented and discussed, leading to meaningful increase of the prediction accuracy of image quality for color distortions.

Edge-Based Robust Image Registration for Incomplete and Partly Erroneous Data

In image registration it is vital to perform matching of those points in a pair of images which actually match each other, and to postpone those which do not match. It is not always known in advance, however, which points have their counterparts, and where are they located. To overcome this, we propose to use the Hausdorff distance function modified by using a voting scheme as a fitting quality function. This known function performs very well in guiding the matching process and supports stable matches even for low quality data. It also makes it possible to speed up the algorithms in various ways. An application to accuracy assessment of oncological radiotherapy is presented. Low contrast of images used to perform this task makes this application a challenging test.

Fuzzy Hough transform-based methods for extraction and measurements of single trees in large-volume 3D terrestrial LIDAR data

This startup study suggests that more accurate and quicker methods of forestry terrestrial LIDAR data analysis can be developed, but new benchmark data sets with the ground truth data known are necessary for these methods to be validated. It follows from the literature review that the improvement in the methods can be attained by the use of newer Hough transform-based (HT) and other robust fuzzy methods for data segmentation and tree measurements. Segmentation of trees can be done by the limit fuzzification of the data around the breast height. Several HT variants having different properties can be applied to measure the diameter at breast height and the accuracies better than those offered by the commercial software seem to be attainable.

Specification of the Evidence Accumulation-Based Line Detection Algorithm

The recently proposed algorithm, using the evidence accumulation principle, for finding lines (ridges) having shape which can be neither parameterized nor tabularized is described in detail. This fuzzy, multi-scale algorithm stores the evidence in the accumulator congruent with the image domain. The primary application was finding blood vessels in mammograms.

Hough Transform for Opaque Circles Measured from Outside and Fuzzy Voting For and Against

Geometrical limitations on the voting process in the classical Hough transform resulting from that the detected objects are opaque to the applied means of measurement are considered. It is assumed that the measurements are made from one point, like in LIDAR scanning. The detected object is ai¾?circle and the two point elementary voting set forming its chord is considered. The first type of conditions are those which can be used during the accumulation process. The side condition says that the circle lies at the opposite side of the chord than the laser source. The magnitude condition requires that points in the elementary set must not be occluded with respect to the source by any circle hypothesised in voting. The second type of conditions can be checked after after the detection. They require that points are neither inside the object not in its shadow. Departures from this rule are admitted, so fuzzy voting between positive and negative evidence for the object is considered.

Finding Line Segments in the Ulam Square with the Hough Transform

The regularities present in the Ulam spiral provided an incentive for interesting observations in the number theory. Therefore, we have made the Ulam square an object of analysis from the image processing perspective. A version of the Hough transform designed specially for detecting sequences of pixels forming segments of straight lines with the slope defined by an irreducible fraction was used to find line segments in the Ulam spiral. Angles which described the slopes of the segments had tangents p / q expressed by integers p from 0 to 10 and q from \(-10\) to 10 (0 excluded). Due to storage limitations the squares with the side of length up to 5001 points which correspond to the largest prime \(25\,009\,991\) were analyzed at present. In such a square the longest segment has 16 primes and its tangent is 3 (3 up and 1 to the right). Segments of length 14 and 15 were absent. The number of shorter segments varied strongly, from one for a 13-point segment to tens of thousands for shorter ones.

Choice of the Hough transform for image registration

Image registration algorithms should be robust against partly erroneous and inconsistent data. The evidence accumulation mechanism known as the Hough Transform (HT) finds the solution indicated by the largest consistent subset of the data. The important case of feature-based registration under the simplified affine transformation, that is, translation, rotation and isotropic scaling, can be easily stated in the terms of HT. Until recently, the use of HT in the considered application was prohibited by excessive computational requirements, but the development of the hardware permanently relieves these limitations. Three versions of the HT, both in the crisp and fuzzy version, were examined against the test images: the Generalized HT (GHT), the Modified Iterated HT (MIHT), and the version called here the Direct Accumulation HT (DAHT), known also as GIPSC on the grounds of photogrammetry. The results indicate that the fuzzy DAHT is robust for over 50% of errors in data, fuzzy GHT up to nearly 30%, and that all the crisp versions as well as the fuzzy MIHT are fragile at least for some examples. The practical applicability of the DAHT and GHT is shown for hierarchical registration of simulation and portal images used in quality assessment of oncological radiotherapy.

The CRASH Project: Defect Detection and Classification in Ferrite Cores

The paper presents a work developed in the framework of the two years COPERNICUS technological research project CRASH (CRack and SHape defect detection in ferrite cores) CIPA-CT94 0I53, in progress since 1995.

Defect Detection in Furniture Elements with the Hough Transform Applied to 3D Data

Defects in furniture elements were detected using data from a commercially available structured light 3D scanner. Out-of-plane deviations down to 0.15 mm were analyzed successfully. The hierarchical, iterated version of the Hough transform was used. The calculation of position of the plane could be separated from that of its direction due to the assumption of nearly horizontal location of the plane, which is natural when the tested elements lie on a horizontal surface.

A Study on Directionality in the Ulam Square with the Use of the Hough Transform

A version of the Hough transform in which the direction of the line is represented by a pair of co-prime numbers has been used to investigate the directional properties of the Ulam spiral. The method reveals the detailed information on the intensities of the lines which can be found in the square and on the numbers of primes contained in these lines. This makes it possible to make quantitative assessments related to the lines. The analysis, among others, confirms the known observation that one of the diagonal directions is more populated with lines than the other one. The results are compared to those made for a square containing randomly located points with a density close to that for the Ulam square of a corresponding size. Besides its randomness, such square also has a directional structure resulting from the square shape of the pixel lattice. This structure does not depend significantly on the size of the square. The analysis reveals that the directional structure of the Ulam square is both quantitatively and qualitatively different from that of a random square. Larger density of lines in the Ulam square along one of the diagonal directions in comparison to the other one is confirmed.