UNSW Canberra

Publication Search Results

Sort
Results per page
Now showing 1 - 1 of 1
  • Light Field Imaging and Reconstruction Tools for Small Objects
    (2024) Sathirasethawong, Chawin
    Thesis
    Inspire by the idea of cataloguing of insects, an automated 3D digitizing system for small objects is introduced in this thesis. Previous existing systems have complicated equipment and time-consuming image acquisition processes which increase the risk of subject damages. As a possible remedy, the light field imaging concept is introduced for small object scanning. Light field imaging has the unique feature of capturing multiple depths of field in one shot, thus reducing acquisition time, and simplifying the required hardware. In small object photogrammetry, the use of a rotary stage decreases the acquisition setting’s complexity. While the setting is simpler, image masking is crucial for the separation between the object and the background. Although there are many masking algorithms in the literature, they have limited performance with 2D images. With the additional depth cue present in light field images, novel object masking algorithms specific to light field images are proposed. The techniques can extract the object of interest from the background by density-based spatial clustering of applications with noise, together with morphological filtering. The result shows that the algorithm works well in varieties of background environment. Moreover, by utilizing the information embedded in the light field images, the algorithms are capable of projecting the mask images from the centre sub-aperture image to adjacent views. Reconstruction of the models is done by implanting photogrammetry. With light field slope information that is obtained from our masking algorithm, the depth specific light field feature extraction algorithm is developed. Without utilising the masks in image preprocessing, the algorithm assigns the depth parameters used in light field feature extraction automatically which speeds up the processing time and also overcomes mistaken features from the background. Throughout this thesis, we introduce the automated 3D digitizing system of small objects using light field technique as a core. The system is flexible and requires less acquisition time for objects that need macrophotography. Moreover, the novel unsupervised object masking algorithm is developed. The developed masking algorithm is promising and helpful especially when there are hundreds of input images. Lastly, the depth specific light field feature extraction algorithm is developed which provides faster processing time and rules out the background even without relying on the masks.