Marko Subasic

Face image validation system

In this paper, we present a novel face image validation system. The purpose of the system is to evaluate quality of face images for identification documents and to detect face images that do not satisfy the image quality requirements. To determine image quality the system first performs face detection in order to find facial features and determine image background. The system consists of seventeen separate tests. Each test checks one quality aspect of the face or of the whole image and compares it to the requirements of International Civil Aviation Organization (ICAO) proposals for machine readable travel documents. The requirements are designed to ensure good conditions for automatic face recognition. The tests are organized in a hierarchical way so the low-level tests are executed first and the high-level tests are executed last. The result of a test is a fuzzy value representing a measure of the image quality. Each test has a set of parameters that can be tuned to produce desired performance of the test. Initial testing of the system has been performed on the set of 190 face images and has demonstrated the feasibility of the method.

Model-based quantitative AAA image analysis using a priori knowledge

Abdominal aortic aneurysm (AAA) is a serious vascular disease which may have a fatal outcome. AAA shape and size is important for diagnostics and intervention planning. In this paper, we present a new method for segmentation of AAA from computed tomography (CT) angiography images. The method works by segmenting the inner and the outer aortic border. Segmentation of AAA is a challenging problem because of low contrast of the outer aortic border. In our method, the inner aortic border is segmented using a geometric deformable model (GDM) and morphological postprocessing. The GDM is implemented using the level-set algorithm. The outer aortic border is segmented by a preprocessing method utilizing a priori knowledge about the aorta shape, followed by the GDM-based method, and morphological postprocessing. The preprocessing algorithm operates on a slice-by-slice basis with some information flow among neighboring slices. The GDM performs three-dimensional (3D) segmentation, reducing possible errors in the previous step. The proposed method is automatic and requires minimal user assistance. The method was statistically validated on 12 patient scans having a total number of 497 image slices. Statistical analysis has confirmed high correlation between the results obtained by the proposed method and the gold standard obtained by manual segmentation by an expert radiologist.

3-D deformable model for aortic aneurysm segmentation from CT images

For treatment of abdominal aortic aneurysm (AAA) by placement of aortic stent graft device it is necessary to make accurate AAA measurements in order to choose the stent graft device of appropriate shape and size. Here, the authors propose a novel technique for 3-D segmentation of abdominal aortic aneurysm from computed tomography (CT) angiography images. The technique is based on 3-D deformable model and utilizes the level-set algorithm for implementation of the method. The method performs 3-D segmentation of CT images and extracts a 3-D AAA model. Once the 3-D model of AAA is available it is easy to perform all required measurements for appropriate stent graft selection. The method proposed here uses the level-set algorithm instead of the classical active contour algorithm developed by Kass et al. (1987). The main advantage of the level set algorithm is that it enables easy segmentation of complex structures such as bifurcations in arteries. In the level set approach for shape modeling, a 3-D surface is represented by a real 3-D function that can be viewed as a 4-D surface. The 4-D surface evolves through an iterative process of solving the differential equation of surface motion. The surface motion is defined by velocity at each point. The velocity is a sum of a constant velocity (inflation force), curvature-dependent velocity (internal force), and image-dependent velocity (external force). The image-dependent velocity is computed on the basis of image gradient. The algorithm has been implemented in MATLAB and C languages. Experiments have been performed using real patient CT angiography images and have shown good results. A 3-D rendering of the segmented region is performed that is useful for aneurysm shape visualization.

Detection of roadside vegetation using features from the visible spectrum

Detection of vegetation in images is a common procedure in remote sensing and is commonly applied to satellite and aerial images. Recently it has been applied to images recorded from within ground vehicles for autonomous navigation in outdoor environments. In this paper we present a method for roadside vegetation detection intended for traffic safety and infrastructure maintenance. While many published methods for vegetation detection are using Near Infrared images which are particularly suitable for vegetation detection, our method uses image features from the visible spectrum allowing the use of common onboard color cameras. Our feature set consists of color features and texture features. One of our specific goals was to identify a useful texture feature set for the problem of vegetation detection. Based on the feature set, the detection is implemented using a Support Vector Machine algorithm. For training and testing purposes we recorded our own image database consisting of different images containing roadside vegetation in various conditions. We are presenting promising experimental results and a discussion of specific problems experienced or expected in real-world application of the method.

Diabetic retinopathy image database(DRiDB): A new database for diabetic retinopathy screening programs research

Diabetic retinopathy is one of the leading disabling chronic diseases, and one of the leading causes of preventable blindness in the world. Early diagnosis of diabetic retinopathy enables timely treatment and in order to achieve it a major effort will have to be invested into screening programs and especially into automated screening programs. For automated screening programs to work robustly a representative fundus image database is required. In this paper we give an overview of currently available databases and present a new diabetic retinopathy database. Our database is to our knowledge the first and only database which has diabetic retinopathy pathologies and major fundus structures annotated for every image from the database which makes it perfect for design and evaluation of currently available and new image processing algorithms for early detection of diabetic retinopathy using color fundus images.

Expert system segmentation of face images

Robust image analysis of photographs for personal documents has been an important open research problem for many years and the interest has been increased by introduction of electronic personal documents, which contain personal digital photographs. International Civil Aviation Organization (ICAO) has defined a set of recommendations defining minimal quality requirements that personal photographs stored in electronic personal documents must satisfy. Some image quality requirements apply only to certain image regions so exact position and location of the image regions has to be known in advance. In this paper, we propose a new knowledge-based method for segmentation of color personal photographs into five regions: skin, hair, shoulders, background, and padding frame. Prior to application of our method, the input image has to be normalized so that both eyes of a person are at the predefined positions within the image. To the best of our knowledge, no method for analysis of personal document photographs has been published in the literature that performs such segmentation. The proposed method consists of two main steps: (i) mean-shift segmentation step; and (ii) region labeling step based on a rule-based expert system. The most important component of the system is a set of rules specifically developed to enable robust labeling of personal document image regions. Extensive experimental validation has been conducted on four image sets and has demonstrated the accuracy and robustness of the proposed method.

3D image analysis of abdominal aortic aneurysm

This paper presents a method for 3-D segmentation of abdominal aortic aneurysm from computed tomography angiography images. The proposed method is automatic and requires minimal user assistance. Segmentation is performed in two steps. First inner and then outer aortic border is segmented. Those two steps are different due to different image conditions on two aortic borders. Outputs of these two segmentations give a complete 3-D model of abdominal aorta. Such a 3-D model is used in measurements of aneurysm area. The deformable model is implemented using the level-set algorithm due to its ability to describe complex shapes in natural manner which frequently occur in pathology. In segmentation of outer aortic boundary we introduced some knowledge based preprocessing to enhance and reconstruct low contrast aortic boundary. The method has been implemented in IDL and C languages. Experiments have been performed using real patient CTA images and have shown good results.

Motion estimation aided detection of roadside vegetation

In this paper we present a method for roadside vegetation detection from video obtained from a moving vehicle with intended use in road infrastructure maintenance and traffic safety. While many published methods are using Near Infrared images which are suitable for vegetation detection, our method uses image features from the visible spectrum allowing the use of a common color camera. The presented detection method uses a set of carefully selected color and texture features. Texture features are based on two-dimensional Continuous Wavelet Transform with oriented wavelets. As selected features can vary with the distance from the camera, we are limiting detection to the regions near to the camera. We used an optical flow algorithm as an approximate estimator of the distance. The classification into vegetation and non-vegetation regions was done using nonlinear SVM. For training and testing purposes we recorded our own video database which contains roadside vegetation in various conditions. We are presenting promising experimental results, comparison with an alternative approach and a discussion of specific problems experienced or expected in real-world application of the method.

Region-based deformable model for aortic wall segmentation

We present a method for automated segmentation of aortic wall from computed tomography angiography (CTA) images, which is the most critical step in quantitative image analysis of abdominal aortic aneurysm (AAA). The two-step method uses geometric deformable models based on the level set algorithm. In the first step, the inner aortic boundary is segmented, followed by segmentation of the outer aortic boundary in the second step. The inner aortic boundary in CTA images have high-contrast edges, which makes this step relatively easy. Segmentation is performed by a 2-D edge-based deformable model. The outer aortic boundary has low contrast edges, resulting in difficult segmentation. To address the problem of ill-defined edges, in our approach we use a 2-D region-based deformable model. We propose several knowledge-based constraints that help in aortic wall segmentation. In our previous research on AAA segmentation, we developed several segmentation methods using edge-based geometric deformable model approaches. In this paper, we present a new region-based segmentation method that uses a geometric deformable model. All specific constraints to the AAA segmentation are incorporated directly into the geometric deformable model, producing a more compact solution. The method has been tested on CTA scans of twelve patients and has shown promising results.

CWT-based detection of roadside vegetation aided by motion estimation

In this paper we present a method for roadside vegetation detection intended for traffic safety and road infrastructure maintenance. While many published methods are using Near Infrared images which are suitable for vegetation detection, our method uses features from the visible spectrum allowing the use of a common color camera. The presented method uses a set of carefully selected color and texture features. Texture features are based on two-dimensional Continuous Wavelet Transform with oriented wavelets. Because texture can vary as the distance from the camera varies, we limit detection to the regions closer to the camera. We use optical flow as an approximate estimator of distance. The classification is done using nonlinear SVM. For training and testing purposes we recorded our own video database which contains roadside vegetation in various conditions. We present promising experimental results as well as a comparison with several alternative approaches.