Nicolae Duta

Matching of palmprints

This paper investigates the feasibility of person identification based on feature points extracted from palmprint images. Our approach first extracts a set of feature points along the prominent palm lines (and the associated line orientation) from a given palmprint image. Next we decide if two palmprints belong to the same hand by computing a matching score between the corresponding sets of feature points of the two palmprints. The two sets of feature points/ orientations are matched using our previously developed point matching technique which takes into account the non-linear deformations as well as the outlier points present in the two sets. The estimates of the matching score distributions for the genuine and imposter sets of palm pairs showed that palmprints have a good discrimination power. The overlap between the genuine and imposter distributions was found to be about 5%. Our preliminary results indicate that adding palmprint information may improve the identity verification provided by fingerprints in cases where fingerprint images cannot be properly acquired (e.g., due to dry skin).

Segmentation and interpretation of MR brain images. An improved active shape model

This paper reports a novel method for fully automated segmentation that is based on description of shape and its variation using point distribution models (PDMs). An improvement of the active shape procedure introduced by Cootes and Taylor (1997) to find new examples of previously learned shapes using PDMs is presented. The new method for segmentation and interpretation of deep neuroanatomic structures such as thalamus, putamen, ventricular system, etc. incorporates a priori knowledge about shapes of the neuroanatomic structures to provide their robust segmentation and labeling in magnetic resonance (MR) brain images. The method was trained in eight MR brain images and tested in 19 brain images by comparison to observer-defined independent standards. Neuroanatomic structures in all testing images were successfully identified. Computer-identified and observer-defined neuroanatomic structures agreed well. The average labeling error was 7%/spl plusmn/3%. Border positioning errors were quite small, with the average border positioning error of 0.8/spl plusmn/0.1 pixels in 256/spl times/256 MR images. The presented method was specifically developed for segmentation of neuroanatomic structures in MR brain images. However, it is generally applicable to virtually any task involving deformable shape analysis.

Less developed corpus callosum in dyslexic subjects—a structural MRI study

BACKGROUND Based on previous studies and due to the characteristics of dyslexia as an auditory phonological decoding disorder, we predicted that the shape of the posterior corpus callosum (CC) would differ between dyslexic and control subjects. METHOD Twenty right-handed boys with developmental dyslexia were selected from a carefully screened general population sample (mean age 11 years) and compared to a matched control group. The CC contour was manually traced on the aligned midsagittal MR slice and total callosal area and its subregions were compared between the groups. A statistical shape analysis and subsequent CC classification was performed using a recently developed shape model method. RESULTS The shape analysis revealed shorter CC shape in the dyslexic group, localised in the posterior midbody/isthmus region. This region contains interhemispheric fibers from primary and secondary auditory cortices. A shape length difference larger than a fixed threshold in the posterior midbody region could correctly discriminate between control and dyslexic subject in 78% of the cases, where a dyslexic CC was shorter in this region than a control CC. However, there were no significant group differences with respect to overall CC area or subregions. CONCLUSION A clear shape difference in the posterior midbody of the CC was found between dyslexic and control subjects. This fits with recent other studies that have reported a strong growth factor in this CC region during the late childhood years, coinciding with literacy acquisition. Our results show that the dyslexic group has not undergone the same growth pattern as the normal reading group.

Deformable matching of hand shapes for user verification

We present a method for personal authentication based on deformable matching of hand shapes. Authentication systems are already employed in domains that require some sort of user verification. Unlike previous methods on hand shape based verification, our method aligns the hand shapes before extracting a feature set. We also base the verification decision on the shape distance which is automatically computed during the alignment stage. The shape distance proves to be a more reliable classification criterion than the handcrafted feature sets used by previous systems. Our verification system attained a high level of accuracy: 96.5% genuine accept rate vs. false accept rate. This performance is further improved by learning an enrolment template shape for each user.

Learning 2D shape models

A new fully automated shape learning method is presented. It is based on clustering a set of training shapes in the original shape space (defined by the coordinates of the contour points) and performing a Procrustes analysis on each cluster to obtain cluster prototypes and information about shape variation. The main difference from previously reported methods is that the training set is first automatically clustered and those shapes considered to be outliers are discarded. The second difference is in the manner in which registered sets of points are extracted from each shape contour. As a direct application of our shape learning method, an 11-structure shape model of brain substructures was extracted from MR image data, an eigen-shape model was automatically trained, and employed to segment several MR brain images not present in the shape-training set. A quantitative analysis of our shape registration approach, within the main cluster of each structure, shows that our results compare very well to those achieved by manual registration; achieving an average rms error of about 1 pixel. Our approach can serve as a fully automated substitute to the tedious and time-consuming manual shape registration and analysis.

Segmentation of the left ventricle in cardiac MR images

This paper describes a segmentation technique to automatically extract the myocardium in 4D cardiac MR images for quantitative cardiac analysis and the diagnosis of patients. Three different modules are presented. The automatic localization algorithm is able to approximately locate the left ventricle in an image using a maximum discrimination technique. Then, the local deformation algorithm can deform active contours so that they align to the edges in the image to produce the desired outlining of the myocardium. Finally, the global localization algorithm is able to propagate segmented contours from one image in the data set to all the others. We have experimented with the proposed method on a large number of patients and present some examples to show the strengths and pitfalls of our algorithm.

Learning Shape Models from Examples Using Automatic Shape Clustering and Procrustes Analysis

A new fully automated shape learning method is presented. It is based on clustering a shape training set in the original shape space and performing a Procrustes analysis on each cluster to obtain a cluster prototype and information about shape variation. As a direct application of our shape learning method, a 17-structure shape model of brain substructures was computed from MR image data, an eigen-shape model was automatically derived. Our approach can serve as an automated substitute to the tedious and time-consuming manual shape analysis.

Learning-based object detection in cardiac MR images

An automated method for left ventricle detection in MR cardiac images is presented. Ventricle detection is the first step in a fully automated segmentation system used to compute volumetric information about the heart. Our method is based on learning the gray level appearance of the ventricle by maximizing the discrimination between positive and negative examples in a training set. The main differences from previously reported methods are feature definition and solution to the optimization problem involved in the learning process. Our method was trained on a set of 1,350 MR cardiac images from which 101,250 positive examples and 123,096 negative examples were generated. The detection results on a test set of 887 different images demonstrate an excellent performance: 98% detection rate, a false alarm rate of 0.05% of the number of windows analyzed (10 false alarms per image) and a detection time of 2 seconds per 256/spl times/256 image on a Sun Ultra 10 for an 8-scale search. The false alarms ore eventually eliminated by a position/scale consistency check along all the images that represent the same anatomical slice.

Road detection in panchromatic SPOT satellite images

The goal of this study is to detect the main road network in high resolution, panchromatic SPOT satellite images. We describe an automatic procedure for road detection which has the following advantages over the previous approaches: 1) it does not require manual initialization; 2) it is able to detect some of the secondary roads in addition to the main highways; and 3) the detection time is small (/spl sim/3 min) even on large images.

Learning the human face concept in black and white images

Presents a learning approach for the face detection problem. The problem can be stated as follows: given an arbitrary black and white, still image, find the location and size of every human face it contains. Numerous applications of automatic face detection have attracted considerable interest in this problem, but no present face detection system is completely satisfactory from the point of view of detection rate, false alarm rate and detection time. We describe an inductive learning-based detection method that produces a maximally specific hypothesis consistent with the training data. Three different sets of features were considered for defining the concept of a human face. The performance achieved is as follows: 85% detection rate, a false alarm rate of 0.04% of the number of windows analyzed and 1 minute detection table for a 320/spl times/240 image on a Sun Ultrasparc 1.