Improving Skin Lesion Segmentation with Self-Training.

Skin lesion segmentation plays a key role in the diagnosis of skin cancer; it can be a component in both traditional algorithms and end-to-end approaches. The quality of segmentation directly impacts the accuracy of classification; however, attaining optimal segmentation necessitates a substantial amount of labeled data. Semi-supervised learning allows for employing unlabeled data to enhance the results of the machine learning model. In the case of medical image segmentation, acquiring detailed annotation is time-consuming and costly and requires skilled individuals so the utilization of unlabeled data allows for a significant mitigation of manual segmentation efforts. This study proposes a novel approach to semi-supervised skin lesion segmentation using self-training with a Noisy Student. This approach allows for utilizing large amounts of available unlabeled images. It consists of four steps-first, training the teacher model on labeled data only, then generating pseudo-labels with the teacher model, training the student model on both labeled and pseudo-labeled data, and lastly, training the student* model on pseudo-labels generated with the student model. In this work, we implemented DeepLabV3 architecture as both teacher and student models. As a final result, we achieved a mIoU of 88.0% on the ISIC 2018 dataset and a mIoU of 87.54% on the PH2 dataset. The evaluation of the proposed approach shows that Noisy Student training improves the segmentation performance of neural networks in a skin lesion segmentation task while using only small amounts of labeled data.

Metrological Evaluation of the Demosaicking Effect on Colour Digital Image Correlation with Application in Monitoring of Paintings.

A modified 3D colour digital image correlation method (3D cDIC) is proposed for efficient displacement measurements of colour objects with natural texture. The method is using a separate analysis of correlation coefficient (sigma) value in the RGB channels of CCD cameras by utilising local information from the channel with the minimum sigma. In this way, merged U, V and W displacement maps are generated based on the local correlation quality. As the proposed method applies to colour filter array cameras, the images in RGB channels have to undergo a demosaicking procedure which directly influences the accuracy of displacement measurements. In the paper, the best performing demosaicking methods are selected. The metrological analysis of their influence on the results of canvas paintings investigations obtained by unmodified and modified 3D cDIC processing is presented.

Multiwavefront digital holographic television.

This paper presents the full technology chain supporting wide angle digital holographic television from holographic capture of real world objects/scenes to holographic display with an extended viewing angle. The data are captured with multiple CCD cameras located around an object. The display system is based on multiple tilted spatial light modulators (SLMs) arranged in a circular configuration. The capture-display system is linked by a holographic data processing module, which allows for significant decoupling of the capture and display systems. The presented experimental results, based on the reconstruction of real world, variable in time scenes, illustrates imaging dynamics, viewing angle and quality.

Wide angle holographic display system with spatiotemporal multiplexing.

This paper presents a wide angle holographic display system with extended viewing angle in both horizontal and vertical directions. The display is constructed from six spatial light modulators (SLM) arranged on a circle and an additional SLM used for spatiotemporal multiplexing and a viewing angle extension in two perpendicular directions. The additional SLM, that is synchronized with the SLMs on the circle is placed in the image space. This method increases effective space bandwidth product of display system data from 12.4 to 50 megapixels. The software solution based on three Nvidia graphic cards is developed and implemented in order to achieve fast and synchronized displaying. The experiments presented for both synthetic and real 3D data prove the possibility to view binocularly having good quality images reconstructed in full FoV of the display.