Ionut Pirnog

Automatic lesion segmentation for melanoma diagnostics in macroscopic images

Detailed segmentation of pigmented skin lesions is an important requirement in computer aided applications for melanoma assessment. In particular, accurate segmentation is necessary for image-guided evaluation of skin lesions characteristics. In this paper, we present a new approach of histogram thresholding for detailed segmentation of skin lesions based on histogram analysis of the saturation color component in the hue-saturation-value (HSV) color space. The proposed technique is specifically developed with the aim to handle the complex variability of features for macroscopic color images taken in uncontrolled environment. A dataset of 30 cases with manual segmentation was used for evaluation. We compare our results with two of most important existing segmentation techniques. For similarity report between automatic and manual segmentation we used dice similarity coefficient (DSC), the true detection rate (TDR), and the false positive rate (FPR). Experimental results show that the proposed method has high precision and low computational complexity.

Image Content Extraction Using a Bottom-Up Visual Attention Model

In this paper we propose a perceptual approach to content analysis based on region segmentation. Content adaptation has become one of the most important problems in recent years due to fast growing of multimedia based services. The process of content analysis and extraction represents one step in solving the problem of content adaptation. Adaptation means the preparation and delivery of content that matches the resources of the connected terminal or network in an optimal way. The proposed content extraction method it can be classified as perceptual because it uses HVS to detect salient regions in images. The main idea of this approach is to adapt the content to the user resources without loss of salient information. Region segmentation is used for object detection and combined with the bottom-up visual attention model can lead to salient object detection.

A Perceptual Approach for Noise Reduction using Nonlinear Spectral Subtraction

We propose an improved spectral subtraction method for reducing acoustic noise added to speech in noisy environments like helicopter cockpit or car engine. This implementation uses over-subtraction method for spectral subtraction. Residual noise can be masked by exploiting the masking properties of the human auditory system. A psychoacoustically motivated weighting filter was included to eliminate residual musical noise. Simulations showed a better quality with no distortion for the enhanced speech and the musical noise was effectively masked.

Improved Minkowsky Metric for Image Region Partition

In this paper we present an improved method for image partition. Image partition plays an important role in video segmentation techniques; it uses different region growth algorithms for separating an image into regions and objects. This enhances the characterization on an image using video descriptors and can be used in image retrieval or image comparison applications. The proposed segmentation method uses the Minkowsky metric for region detection. In addition, it uses different threshold values in order to achieve perceptually significant areas. The simulation results indicate good performances of the proposed method for colour image segmentation.

Efficient Method for Content Extraction Applied in Multimedia Communication

In this paper we present the multimedia content extraction method based on region and object segmentation. The process of content extraction represents one step in solving the problem of content adaptation. Adaptation means the preparation and delivery of content that matches the resources of the connected terminal or network in an optimal way. The process of multimedia content extraction consists of two stages: semantic video modeling and video segmentation. The result of semantic video modeling is the representation of raw data in a more structured form, and it is essential in the following stage. Video segmentation means the partition of an image into a set of non overlapping homogenous regions whose union is the entire image. The presented segmentation methods are edge-based, region-based and motion-based, and are used for moving or static object detection.

Cutaneous melanoma risk evaluation through digital image processing

Melanoma is the most aggressive type of human cancer but also the most likely to be cured if detected in its early stages. The risk of skin lesions to turn into melanoma can be easily evaluated by dermatologists using a non-invasive procedure called dermoscopy. This paper addresses the issue of melanoma risk evaluation through digital processing of macroscopic skin lesions images. The main goal of our research is the development of an application for tracking the time evolution of skin lesions and evaluating the risk of these lesions to turn into melanoma. The first step in the evaluation process is the extraction of the pigmented skin lesions by means of image segmentation. The two objectives presented in this paper are the determination of the best suited segmentation method in terms of computational complexity and the evaluation of the color space used for digital image representation in order to obtain good initial segmentation results.

New Content Analysis Methods Effective in Multimedia Based Services

In this paper we present new content analysis methods effective in multimedia based services. The problem of multimedia content delivery and retrieval has become more and more important due to the growing amount of video information that is delivered to consumers today. The classical searching and retrieval mode in multimedia databases is by name; however the name cannot truly describe the semantic content of most video content. The solution for this problem is visual content extraction. This means that the multimedia content will be characterized not only by name but also by the most representative information contained. This information is known as visual feature and there are any methods that try to extract visual features from multimedia streams. The methods presented in this paper are based on colour and spatial information and are used for extracting patterns from images and videos. The process of multimedia content retrieval consists of three stages: semantic video modelling, video segmentation and feature extraction. The result of semantic video modelling is the representation of raw data in a more structured form, and it is essential in the following stage. Video segmentation means the partition of an image into a set of non overlapping homogenous regions whose union is the entire image. The presented segmentation methods used are edge-based, region-based and motion-based, and are used for moving or static object detection.

Digital watermarking and segmentation of macroscopic pigmented skin lesions images

This work presents the use of digital watermarking for applications of automated pigmented skin lesions assessment. Automated skin lesion assessment by digital image processing of macroscopic pigmented skin lesion (MPLS) images uses lesion characteristics for malignant risk computation. The accuracy of the computed risk level is increased when additional information is taken into consideration: age, gender, skin type, position of the lesions on body, and the presence of multiple and similar lesions. We propose a digital watermarking scheme for embedding relevant patient information into the color MPSL image. The preliminary experimental results are promising, and suggest that the proposed watermarking scheme is suitable for this type of application.

Efficient Transform Coefficient Coding in HEVC

This paper presents an efficient coding method of transform coefficients in HEVC video encoding standard. Our approach makes use of spectral characteristics of the transform unit coefficients in inter predicted coding units. Also a coefficient significance assessment algorithm is integrated into the video encoder, for a non-zero coefficients number reduction. It is shown to be a good method from the compression point of view, without quality losses.

Multiple point Android digital image authentication

Digital images have been accepted as evidence in courts of law for a long time. One mandatory aspect for the admission of photographic evidence is the need for authentication. A spontaneous scene is very likely to be photographed with a smartphone rather than a classical camera. Android powered devices count for around 80% of smartphones worldwide. There is a high probability that an image used as evidence has been captured with this type of device. Android uses JPEG format for storing captured images. JPEG headers embedded information exhibits a wide variation across different models and manufacturers. By taking advantage of this variation, the authentication method described in this article employs a multiple point verification of both intrinsic and extrinsic properties of the analyzed image. The extrinsic properties are being obtained from a 185 entries database of Android powered devices. Authentication is performed straight-forward, allowing the authenticity decision to be made at different points in the process, without the need of a full analysis to be employed beforehand.