Contribuţii la dezvoltarea senzorilor de imagine inteligenți / Contributions to the development of computational imaging sensors

With the development and mass production of spatial light modulators, alternative architectures for optical sensors have gained interest. These fall under the umbrella of computational imaging where the desired signal is not measured directly by sensors but related measurements are taken and the signal is reconstructed from them. With this approach the signal can be reconstructed with less noise than in the classical approach for certain cases like far-infrared and terahertz imaging. Additionally, with better models of the underlying signal or the sensor architecture further improvements can be made in the accuracy of acquisition. In this thesis we analyze Hadamard spectroscopy and single-pixel imaging architectures. We study the noise reduction advantage of Hadamard spectroscopy (Fellget advantage) both theoretically and experimentally and find that the advantage is mitigated by factors related to the light sensor. We analyze the noise reduction effect of micro-scanning and its dependence on the number of previews taken. We develop a model for the formation of the point spread function of a single pixel camera and a method for experimental evaluation. We then apply the method in order to improve the image with deconvolution.