DEEP LEARNING-BASED UNROLLED RECONSTRUCTION METHODS FOR COMPUTATIONAL IMAGING

Download

Thesis_CanDenizBezek.pdf

Date

2021-9-08

Author

Bezek, Can Deniz

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

573
views

571
downloads

Computational imaging is the process of forming images from indirect measurements using computation. In this thesis, we develop deep learning-based unrolled reconstruction methods for various computational imaging modalities. Firstly, we develop two deep learning-based reconstruction methods for diffractive multi-spectral imaging. The first approach is based on plug-and-play regularization with deep denoisers whereas the second one is an end-to-end learned reconstruction based on unrolling. Secondly, we consider general multidimensional imaging systems whose measurements involve convolution and superposition. We formulate a realistic truncated image formation model and develop a deep learning-based unrolled reconstruction method to solve the associated inverse problem. The performance is illustrated for single-frame deconvolution and diffractive multi-spectral imaging applications. Thirdly, we extend the deep learning-based unrolled reconstruction methods to a compressive spectral imaging modality by considering both the conventional and truncated image formation models. The performance of all the developed methods is comparatively evaluated for sample applications using different convolutional neural network (CNN) architectures, including U-Net. Results illustrate the superior performance of the developed unrolled learned reconstruction methods over purely analytical methods for all simulation settings. We also demonstrate the generalization capability of the developed unrolled reconstruction methods through extensive simulations and perform model mismatch analyses to illustrate the improvement gained with the truncated image formation model.

Subject Keywords

computational imaging, spectral imaging, inverse problems, image reconstruction, deep learning, algorithm unrolling, truncated convolution

URI

https://hdl.handle.net/11511/93159

Collections

Graduate School of Natural and Applied Sciences, Thesis

Suggestions

OpenMETU
Core

Anlık Spektral Görüntüleme için Tasarım Eniyileme Ayazgök, Suleyman; Öktem, Sevinç Figen (2019-08-22) Snapshot spectral imaging enables to reconstructspectral images from a multiplexed single-shot measurement.Since an inversion is required to form the spectral images com-putationally, quantitative characterization of their performanceis essential to optimize the design. In this paper, we analyze theoptimal design of a snapshot spectral imaging technique. Thissnapshot multi-spectral imaging technique uses a diffractive lenscalled generalized photon sieve, and vari...
Efficient algorithms for convolutional inverse problems in multidimensional imaging Doğan, Didem; Öktem, Figen S.; Department of Electrical and Electronics Engineering (2020) Computational imaging is the process of indirectly forming images from measurements using image reconstruction algorithms that solve inverse problems. In many inverse problems in multidimensional imaging such as spectral and depth imaging, the measurements are in the form of superimposed convolutions related to the unknown image. In this thesis, we first provide a general formulation for these problems named as convolutional inverse problems, and then develop fast and efficient image reconstruction algorith...
Compressive spectral imaging using diffractive lenses and multi-spectral sensors with learned reconstruction and joint optimization Gündoğan, Utku; Öktem, Sevinç Figen; Department of Electrical and Electronics Engineering (2022-2) Compressive spectral imaging aims to reconstruct the entire three-dimensional spectral cube from a few measurements, ideally with a snapshot capability. Recently various spectral imaging modalities have been developed by exploiting diffractive lenses. Another line of development in this area is enabled by spectral filter arrays which resulted in multi-spectral sensors. In this thesis, we first review an existing compressive spectral imaging modality with diffractive lenses and analyze its performance using ...
End-to-end learned image compression with conditional latent space modelling for entropy coding Yeşilyurt, Aziz Berkay; Kamışlı, Fatih; Department of Electrical and Electronics Engineering (2019) This thesis presents a lossy image compression system based on an end-to-end trainable neural network. Traditional compression algorithms use linear transformation, quantization and entropy coding steps that are designed based on simple models of the data and are aimed to be low complexity. In neural network based image compression methods, the processing steps, such as transformation and entropy coding, are performed using neural networks. The use of neural networks enables transforms or probability models...
Automatic building extraction from high resolution satellite images for map updating: A model based approach San, D. Koc; TÜRKER, MUSTAFA (2007-10-12) An approach was developed for automatically updating the buildings of an existing vector database from high resolution satellite images using spectral image classification, Digital Elevation Models (DEM) and the model-based extraction techniques. First, the areas that contain buildings are detected using spectral image classification and the normalized Digital Surface Model (nDSM). The classified output provides the shapes and the approximate locations of the buildings. However, those buildings that have si...

Citation Formats

C. D. Bezek, “DEEP LEARNING-BASED UNROLLED RECONSTRUCTION METHODS FOR COMPUTATIONAL IMAGING,” M.S. - Master of Science, Middle East Technical University, 2021.