Kamil Riha

Circle detection in pulsative medical video sequence

The article deals with a new method for the detection of pulsative circular objects in a medical video sequence. The motivation for investigating this method consists in the fact that a circular object is not very apparent and its detection in such a frame is inaccurate. In some cases of medical images, the pulsative character of the circular area being searched can be used for its localisation. The proposed method starts from an analysis of movement, using optical flow estimation. The compensation of global movement is necessary because only local pulsative movement during the video sequence is assumed. The optical flow estimation is followed by another main processing step: the Hough Transform for the circle position estimation. Circles with expected properties are selected using the Bayes classifier. Finally, the circle position in a single frame is adapted using the analysis of average pixel intensity in the directions starting from the circle centre.

Object localization in medical images

This paper is focused on localization of objects in medical images. A novel improvement of an existing method for localization of artery in longitudinal ultrasound B-mode scan is proposed in the paper. The localization is based on a classification of pixels according to image information in their neighborhood. To suppress misclassified points, a novel RANSAC based method is proposed. This method is able to find the most appropriate mathematical model of depicted common carotid artery (CCA) on the basis of previous classification. The proposed RANSAC based method with its mathematical footing is described in detail and the results of method within the algorithm for localization of artery are enclosed. By using the proposed RANSAC based method the localization method becomes very robust.

An efficient automatic method of Optic disc segmentation using region growing technique in retinal images

Segmentation of Optic disc (OD) from a retinal image is a essential step while developing automated screening systems for eye disease like diabetic retinopathy, Glaucoma etc. This paper proposes a method of automatic optic disk segmentation based on region growing technique with automatic seed selection. In this method centre of optic disk is considered as a seed to apply region growing technique to segment the optic disk from the preprocessed retinal image. Automatic detection of centre of optic disk is done by double windowing method. The algorithm uses image processing techniques like contrast adjustment, morphological operations & filtering to process the retinal image and to remove the blood vessels from the retinal image. The performance of optic disk segmentation by proposed method is compared with Optic disk segmentation by ophthalmologists and results are found convincing and efficient. The experimental results indicate this method of segmentation of the OD has good accuracy and also is computationally cheap.

Watermark generation from fingerprint features for digital right management control

The conventional digital watermarking schemes uses an arbitrary digital pattern as the watermark which has limitations in proving ownership of the watermark. This paper proposes a proficient digital watermark generation technique from biometric data which will be unique and can be logically owned to prove ownership. The issue of ownership watermark is addressed in this paper. The biometric pattern of fingerprint is used to generate the digital watermark that has a stamp of ownership. The generated watermark has been studied for uniqueness and identification and has been used to watermark digital images. Discrete cosine transformation is used for embedding the watermark in the image. Experimental results indicate that the watermark can survive the signal processing attacks and maintain the perceptual properties of the host signal. The extracted biometric based watermark was uniquely identified under signal processing attacks by matching of the feature points.

Automatic exudates detection in fundus image using intensity thresholding and morphology

Diabetic retinopathy (DR) is a leading cause of blindness in diabetic patients. Exudates are one of the most common earliest signs of diabetic retinopathy. Automatic and accurate detection of exudates in fundus images is an important step in early diagnosis of DR. In the proposed method detection of exudates, two independent approaches based on intensity thresholding and morphological processing are strategically combined to detect any small exudates present while removing all possible types of false positives. This strategic combination removes the noise sources from blood vessels and reflections during image capture making the detection of exudates accurate. Experimental results indicate that the proposed method has good accuracy in exudates detection without compromising the computational time and hence can be considered for screening purpose of DR.

Automatic glaucoma detection using adaptive threshold based technique in fundus image

Glaucoma is a kind of ocular disorder that results in a damaged optic nerve which is responsible for transmitting images to the brain. The conventional methods to detect glaucoma like Optical Coherence Tomography (OCT) and Heidelberg Retinal Tomography (HRT) are expensive and need specialized manpower. A digital fundus image can be used to identify glaucoma. This paper describes an efficient method to analyze a computer-aided fundus image which can act as a diagnostic tool for detection of glaucoma. A technique based on histogram of the image is used to study some statistical features of the image such as mean and standard deviation. A relationship between them is established to find a threshold value for segmenting optic disc and optic cup. An adaptive threshold based method which is independent of image quality and invariant to noise is used to segment the optic disc, optic cup and the cup-to-disc ratio CDR which is used to screen glaucoma. The experimental results obtained are compared with those of the ophthalmologist and are found to have a high accuracy. Also in addition the proposed method is efficient having a low computational cost.

Segmentation of brain tumor parts in magnetic resonance images

The problem most frequently encountered in the practical processing of medical images consists in the lack of instruments enabling machine evaluation of the images. A typical example of this situation is perfusion analysis of brain tumor types. The first and very significant step lies in the segmentation of individual parts of the brain tumor; after segmentation, the rate of penetration by the applied contrast agent is observed in the parts. The common method, in which a high error rate has to be considered, is to mark these tumor portions manually. Within the second step of the segmentation procedure, the monitoring of perfusion in the segmented tissues is realized together with the correlation to model cases. The quality of brain tissue segmentation exerts significant influence on the quality of evaluation of perfusion parameters; consequently, the tumor type recognition is also influenced. This means that the design of a suitable, accurate, and reproducible method constitutes a critical point within perfusion analysis. In this context, reproducibility is an important aspect owing to the preservation of segmentation conditions in monitoring the development of a tumor in time. The authors describe classification methods enabling the segmentation of images acquired via magnetic resonance tomography.

The method for material corrosion modelling and feature selection with SVM-RFE

Material corrosion has caused more and more losses and costs these years, so the world begin to pay much attention to this problem. In this paper, we mainly discuss the modeling and feature selection of Material corrosion data. With our experimental data with very small sample size, a model of corrosion rate is built. After specialized data preprocessing. By combining RFE and SVM, a novel feature selection method SVM-RFE is introduced. Then integrating this feature selection method and SVM modeling method, a special modeling framework is built. According to the experiments, the priority of this method is established not only on algorithm efficiency but also on predicting precision.

A novel immune image template set for fuzzy image segmentation and its application research

Image segmentation is one of the classic problems in the computer vision field. Although a lot of successful operators and algorithms have been proposed, fuzzy image segmentation does not always achieve satisfactory results. This paper is inspired by Positive Selection Algorithm and Negative Selection Algorithm and, is based on the mechanism and process where T-cell is activated by the MHC molecule. A new positive selection algorithm is introduced which establishes so-called templates set for immune detection. This algorithm is based on processing of image information represented as a gray value statistic rather than arithmetic gradient formulation. It is comprised of a template set not just a single template. Therefore it gives good results for different images. The presented algorithm is used for image segmentation into objects, background and fuzzy edge in fuzzy infrared images.

Cervical cancer detection and classification using Independent Level sets and multi SVMs

Introduced in 1940, Pap smear test has proven to be an effective screening method to determine the different stages of cervical cancer. Identification and classification of Pap smear images to detect cervical cancer via manual screening is a challenging task for pathologists therefore increasing the chances of human error. In this paper, we propose an automatic method to detect and classify the grade of cervical cancer using both geometric and texture features of Pap smear images and classifying accordingly using multi SVM. The geometric features are obtained through segmentation of nucleus and cytoplasm using independent level sets, detecting whether the cell is cancerous or normal, with reference to the ground truth. By extracting well defined GLCM texture features and using a combination of PCA and the best class of multi SVM, the images are classified with an accuracy of 95%.