Algorithms for Visual Computing and Biomedical Image Processing

Abstract

With the advances in computation techniques, the development of computer vision and graphics algorithms, and their connection with people’s lives are getting closer, there has been a strong desire to integrate the computational methodologies into traditional applications and research directions. For example, in three-dimensional display, some core factors, such as motion parallax and binocular disparity, are extremely crucial in creating an immersive display with fidelity and vividness by providing depth cues to the brain during gaining 3D sensations. However, in the past decades, the performances are limited because the former methods fail in building a 3D perception to the users with depth cues in high accuracy. Fortunately, in the modern researches, there are many state-of-art works showing fascinating results through applying sophisticated equipment and developed computational methodologies. And the other fields, such as user interface design, sketching, art, and biological research, also benefit from powerful computational methods and obtain huge enhancement on performance. This dissertation focuses on algorithm investigation and implementation for various application scenarios in advanced manners.

Three computational applications are explored in this dissertation. In the first part, a novel volumetric reach-through display is implemented by projecting objects on a reconfigurable and non-planar fog screens. Two prototypes are proposed: one makes use of a 2D array of individually switchable fog nozzles to construct a fog screen; the other fog display is composed of stand-alone fog units attached on continuous moving linear platforms. The prototypes are designed closely packed to avoid the air turbulence suffered in former works. The contents projected onto the fog screens are tightly synchronised with screen patterns under the control of the developed algorithm and software. The proposed system allows high- resolution 3D images to be observed without tracking unit, special glasses or head-mounted device. Beyond the display function, the mid-air and immaterial fog display superimposes displayed content on the real world and is directly touchable and controllable for users, facilitating Augmented Reality and Human Computer Interaction.

The second application is inspired by the conventional freehand sketch-based light painting. In this dissertation, a more advanced computational light painting method is proposed to create a precise mid-air 3D image through moving a flat display attached to a robot arm and using camera long exposure to record all the light trajectory. An automatic path generation algorithm is developed to yield robot path while accommodating the 3D shapes of objects to achieve a higher space efficiency. All the light contours are rendered in real-time using an Octree-based slicing algorithm to keep the displayed contour synchronised with the robot pose, which significantly improves the performance and reduces the time cost. The extended application of light painting, named as kinetic photography, is also explored using the same computational methods. Instead of keeping the camera stationary, we move the camera with robot and capture long exposures of a stationary display showing light contours.

A combination of computer-aided method and biomedical research is reported in the last part in this dissertation. Cellular chirality is an important research topic in biological studies, which may contribute to the development of left-right asymmetry in tissue/organ formation. It still remains a big challenge for biomedical researchers to analyse a large amount of cellular information and determine the orientation of each cell, especially in a cell cluster. With the developed algorithm, it is possible to segment overlapping nucleus contours from densely-packed clusters and extract the single cell information. The result reveals an enhanced chirality, which is consistently observed with improved intercellular alignment and being independent from cell-cell distance. By segmenting single cells with intact orientation, this automatic method offers a quantitative analysis with improved accuracy, which provides an essential tool for studying left-right asymmetry and other morphogenetic dynamics in tissue regeneration.

Date of Award	9 Jan 2019
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Ting Hsuan CHEN (Supervisor) & Miu Ling LAM (Co-supervisor)