Photo Credi: Antoni Costa / UIB
The 21th October of 2022 Onofre Martorell presented his PhD thesis: Geometrical image characterization and multi-exposure imaging
Title: Geometrical image characterization and multi-exposure imaging
Author: Onofre Martorell
Directors: Dr. Antoni Buades
Date: 21 October 2022
Abstract:
In this thesis we propose novel algorithms and techniques for two different problems in the fields of image processing and computer vision.
In the first part we propose a basic taxonomy of image contours. Our goal is to classify smooth curves into five categories, namely, circles, ellipses, line segments, arcs of circles and arcs of ellipses. These geometrical structures have been chosen as they serve as input of many computer vision tasks. The proposed strategy is applied on a set of initial disjoint contours, which are grouped together to form the aforementioned structures. These, in turn, are validated using an a contrario approach that guarantees a reduced number of false detections. The use of a multiscale strategy permits the detection at different resolution levels, which makes the method robust to noise and blur.
In the second part we propose three methods for the combination of multi-exposure images. First, we propose a novel algorithm for multi-exposure fusion (MEF) which is able to deal with {sequences where there is camera or object motion}. This algorithm decomposes image patches with the Discrete Cosinus Transform (DCT) transform. Coefficients from patches with different exposure are combined. The algorithm adapts to dynamic sequences in order to avoid { the so-called ghosting artifacts, that appear when there are misalignment on the fused images}. Experiments with several data sets show that the proposed algorithm performs better than state-of-the-art. We then take into account the joint MEF and noise removal. Both tasks are performed simultaneously in the DCT domain, which leads to a very efficient algorithm. The method takes advantage of spatio-temporal patch selection and collaborative 3D thresholding. Several experiments show that the obtained results are significantly superior to the existing state-of-the-art.
Finally, we propose a patch-based method for the simultaneous denoising and High Dynamic Range of a sequence of multi-exposed RAW images. A spatio-temporal criterion is used to select similar patches along the sequence, and a weighted principal component analysis both denoises and fuses the multi-exposed data simultaneously. Several experiments show that the proposed method obtains state-of-the-art fusion results with real~RAW~data.
