Chair: Stephen Marshall, University of Strathclyde, UK
Ronald Jones, CSIRO (Australia)
In this paper we present algorithms for non-flat connected component filters using the notion of a component tree. The advantage of using non-flat filters as compared to flat filters is that they can access and utilise linking between components at sequential graylevels in the image. We have found that this information can be used to develop powerful new connected filters with practical applications . One of the key benefits of the approach is that the image features to be filtered undergo the maximum amount of filtering that is possible while leaving the rest of the image untouched. As a consequence, a segmentation of the features can then be obtained simply by locating those pixels within the image that have been changed by the filter.
C. I. Cotsaces, University of Thessaloniki (Greece)
Ioannis Pitas, University of Thessaloniki (Greece)
The influence zone transform is a fundamental tool in morphological and qualitative digital image processing. Because of its inherent geodesic properties, it is most efficiently computed using propagation front or grassfire based methods. However, when the image processed is too large to be contained in available memory, the random access nature of these algorithms makes them exceptionally inefficient. In order to alleviate this problem, we have developed two algorithms that greatly reduce the memory requirements of the transform. The first is designed specifically for computing the influence zone transform on surfaces, without storing the volume enclosing the surface. The second performs the transform using only the propagation fronts, and without storing any part of the region that is being processed. Both methods use much less memory than the ones in the literature, and thus enable the transform to be performed on much larger images than before. However, since all three algorithms use a significant number of set-access operations, they are considerably slower that their classical counterparts. Several techniques have been developed in this work in order to minimize the effect of these set operations. These include fast search methods, double propagation fronts, directional propagation, and others.
Sharmila Kannangara, Purdue University (U.S.A.)
Eduardo Asbun, Purdue University (U.S.A.)
Robert X. Browning, Purdue University (U.S.A.)
Edward J. Delp, Purdue University (U.S.A.)
The recognition of text overlay information appearing in video frames can be used for classification and scene indexing for archival purposes. In this paper, an algorithm for text recognition in C-SPAN images is presented. A method for the segmentation of text blocks into individual letters is outlined. A recognition method using shape sensitive morphological operations is presented.
Pertti Koivisto, Tampere University of Technology (Finland)
Pauli Kuosmanen, Tampere University of Technology (Finland)
A new method of controlling the trade-off between noise attenuation and detail preservation in nonlinear filter design is presented. The technique is based on an appropriate combination of the sample selection probabilities of a filter with traditional error criteria. The practical applicability of the approach is empirically studied in connection with the training-based optimization of soft morphological filters. Also, the formulas for the sample selection probabilities of the basic soft morphological filters are derived.
Xiao Pei, Tampere University of Technology (Finland)
Moncef Gabbouj, Tampere University of Technology (Finland)
Two approaches of multistage gradient robustification for image contour detection are presented in this paper: two stages of Difference of Estimates and Difference of Estimate followed by an optimal filtering. Watershed transformation is then applied to these robustified gradient images to effectively detect image contours which are guaranteed to be in closed form. Multistage gradient robustification provides the flexibility of using different image processing techniques and produces good detection results for the images highly corrupted with noise.
Marcin Iwanoski, Warsaw University of Technology (Poland)
Slawomir Skoneczny, Warsaw University of Technology (Poland)
Jaroslaw Szostakowski., Warsaw University of Technology (Poland)
This paper is devoted to a segmentation method using advanced morphological filtering by reconstruction followed by clustering by k-means algorithm. Advanced morphological filtering bases on morphological reconstruction and two filters are applied: opening by reconstruction and closing by reconstruction. This kind of operation has very important advantage from the point of view of segmentation - it preserves the borders of regions. Traditional filters (opening, closing, linear filters) remove noise, but on the other hand they cause some blur effects, which can be the serious obstacle for correct segmentation. Morphological filtering by reconstruction has very good filtration properties without changing the shapes. After segmentation simple k-means clustering is performed. Two versions of k-means clustering algorithm is described: classic and fuzzy one. First, ?crisp' version will be applied to cases with a knowledge regarding number of clusters given a priori. Fuzzy version should be used when it is difficult to define number of clusters. The algorithm will automatically adapt number of clusters into the structure of the image. A combination of filtering by morphological reconstruction and clustering makes possible to consider two kind of information: spatial (filtering) and spectral (clustering).