Agnieszka Stankiewicz

Improving of Speaker Identification from Mobile Telephone Calls

The paper examines issues related to proper selection of models used for quick speaker recognition based on short recordings of mobile telephone conversations. A knowledge of the encoder type used during the transmission of speech allows to apply an appropriate model that takes specific characteristics of the encoder into account: full rate (FR), half rate (HR), enhanced full rate (EFR) and adaptive multi-rate (AMR). We analyse both proper model selection and automatic silence removal. Analysis of time of processing is also a part of this study.

Novel full-automatic approach for segmentation of epiretinal membrane from 3D OCT images

The scientific objective of the research presented in this paper was to develop a new OCT-based method for investigation of epiretinal membrane (ERM) pathology in human eyes. We propose a new approach for the automatic assessment of the ERM thickness and the space between the ERM and the retina surface. The experiments were conducted using volumetric data acquired from a set of patients with ERM. OCT cross-sections were processed frame-by-frame. The utilized advanced digital imaging algorithms included anisotropic diffusion method for denoising, graph analysis approach for ERM surface segmentation, and region growing and thresholding technique for under-ERM space segmentation. We were able to segment and quantify ERMs in all subjects with good precision. The calculated information is presented in the form of depth maps.

Volumetric segmentation of human eye blood vessels based on OCT images

In this paper we present a method for volumetric segmentation of retinal vessels based on 3D OCT images of human macula. The proposed hybrid method is comprised of two steps: detailed extraction of superficial blood vessels indicators visible in 2D projection of retina layers followed by an axial inspection of inner retina to determine exact depth position of each vessel. The segmentation procedure is improved by application of block-matching and 4D filtering (BM4D) algorithm for noise reduction. The 3D reconstruction of vascular structure was performed for 10 normal subjects examined with Avanti AngioVue OCT device. The automated segmentation results were validated against the manual segmentation performed by an expert giving the accuracy of 95.2%.

Imaging and Measurement of the Preretinal Space in Vitreomacular Adhesion and Vitreomacular Traction by a New Spectral Domain Optical Coherence Tomography Analysis

Purpose: To evaluate a new method for volumetric imaging of the preretinal space (also known as the subhyaloid, subcortical, or retrocortical space) and investigate differences in preretinal space volume in vitreomacular adhesion (VMA) and vitreomacular traction (VMT). Methods: Nine patients with VMA and 13 with VMT were prospectively evaluated. Automatic inner limiting membrane line segmentation, which exploits graph search theory implementation, and posterior cortical vitreous line segmentation were performed on 141 horizontal spectral domain optical coherence tomography B-scans per patient. Vertical distances (depths) between the posterior cortical vitreous and inner limiting membrane lines were calculated for each optical coherence tomography B-scan acquired. The derived distances were merged and visualized as a color depth map that represented the preretinal space between the posterior surface of the hyaloid and the anterior surface of the retina. The early treatment d retinopathy study macular map was overlaid onto final virtual maps, and preretinal space volumes were calculated for each early treatment diabetic retinopathy study map sector. Results: Volumetric maps representing preretinal space volumes were created for each patient in the VMA and VMT groups. Preretinal space volumes were larger in all early treatment diabetic retinopathy study map macular regions in the VMT group compared with those in the VMA group. The differences reached statistical significance in all early treatment diabetic retinopathy study sectors, except for the superior outer macula and temporal outer macula where significance values were P = 0.05 and P = 0.08, respectively. Overall, the relative differences in preretinal space volumes between the VMT and VMA groups varied from 2.7 to 4.3 in inner regions and 1.8 to 2.9 in outer regions. Conclusion: Our study provides evidence of significant differences in preretinal space volume between eyes with VMA and those with VMT. This may be useful not only in the investigation of preretinal space properties in VMA and VMT, but also in other conditions, such as age-related macular degeneration, diabetic retinopathy, and central retinal vein occlusion.

Matching 3D OCT retina images into super-resolution dataset

Optical coherence tomography (OCT) is the current very fast and accurate modality for noninvasive assessment of 3D retinal structure. Due to large amount of data acquired with this technique the resolution of 3D scans is limited. In this paper we present a new method for improving resolution of 3D macula scans while maintaining short acquisition time and robustness with respect to motion artifacts. Our approach is based on multiframe super-resolution method applied to several 3D standard resolution OCT scans. Presented experiments where performed on volumetric data acquired from adult patients with the use of Avanti RTvue device. Each OCT cross-section (B-scan) was subjected to image denoising and retinal layers segmentation. The generated 3D super-resolution scans have significantly improved quality of the vertical cross-sections.

Automatic modeling and classification of vitreomacular traction pathology stages

Retinal pathologies that are detected too late and/or left untreated can seriously damage eyesight. It is important to monitor the retina and react to any pathological changes. A fast, accurate, non-invasive, and even three-dimensional retina examination is the optical coherence tomography (OCT). In this paper we propose a new automated classification method for evaluation of vitreomacular interface (VRI) in human eyes. We present an approach for modelling changes in retina structure during the progression of vitreomacular traction (VMT) pathology. Presented experiments were performed on volumetric data acquired from adult patients with the use of Avanti RTvue device. Advanced digital image processing algorithms were subsequently applied to each OCT cross-section (B-scan) for image denoising and flattening, as well as retina layers segmentation. The proposed solution has a good accuracy and almost all subjects were successfully classified into one of 4 groups corresponding to various stages of VMT. The developed models of VMT stages show a high potential of the proposed method to support ophthalmologists in making appropriate clinical decisions.

Real Time Recognition Of Speakers From Internet Audio Stream

Abstract In this paper we present an automatic speaker recognition technique with the use of the Internet radio lossy (encoded) speech signal streams. We show an influence of the audio encoder (e.g., bitrate) on the speaker model quality. The model of each speaker was calculated with the use of the Gaussian mixture model (GMM) approach. Both the speaker recognition and the further analysis were realized with the use of short utterances to facilitate real time processing. The neighborhoods of the speaker models were analyzed with the use of the ISOMAP algorithm. The experiments were based on four 1-hour public debates with 7–8 speakers (including the moderator), acquired from the Polish radio Internet services. The presented software was developed with the MATLAB environment.