This manuscript is a condensed report of my scientific work in unconventional imaging techniques, including polarimetric imaging, quantum imaging and imaging through turbid media. Ranging from theory, through instrumental system design, to statistical signal processing, my scientific contribution aims at covering, as best as possible, the whole chain of understanding behind a physical imaging problem. This position is in phase with the recently celebrated “co-design” way of thinking which consists of jointly optimizing the hardware (physics of imaging, instrumental setup) and software (signal/image processing) building blocks required to provide the best image quality (of best processing performance) to an end-user for a dedicated application. This work has been conducted over the last ten years, during my PhD at Institut Fresnel, in the Physics and Image Processing group (Phyti), then through a post-doctoral position at ONERA (the french aerospace lab) in the Lidar and Laser Sources group (DOTA-SLS), and finally at the Institute of Physics of Rennes, Optics and Photonics Department, where I actively contributed to create a recognized research activity in imaging (Advanced imaging axis). In an attempt to present a unified vision of my scientific work, this document will be decomposed into two parts, corresponding to the two main aspects of my research activity. However, as imposed by any frontier, many connections between them are to be found, in terms of either applicative or scientific domains, which will be made appearant in this manuscript. The first part, Physics of light depolarization and non conventional polarimetric imaging, is focused on polarimetric imaging, whose principles, methods and main topical challenges will be briefly recalled in Chapter 1. In spite of its long history, and of the huge number of contributors in this domain (either academic or industrial), this unconventional imaging technique still opens interesting scientific perspectives, in terms of physical understanding of the light-matter interaction, especially in random scattering media. As a result, there is still a topical debate about the clear definition and understanding of light depolarization, which is however one of the most used figure of merit in a wide number of polarimetric imaging applications. The theoretical and experimental developments conducted during these last years have allowed us to gain some physical insight in this domain, which will be summarized in Chapter 2. The second challenge that I have tried to tackle in the polarimetric imaging domain concerns the simplification and robustness of the imaging processes. Indeed, full-Stokes or full-Mueller imaging polarimeters are very powerful and precise instruments, but to the expense of sometimes unaffordable price, time-consuming acquisitions, low wavelength tunability and complex calibration, correction and post-processing tasks in order to indirectly retrieve relevant polarimetric parameters. In that context, I will briefly present in Chapter 3 various simplified polarimetric imaging techniques, that aim at strongly reducing acquisition times, cost, and complexity of the image interpretation by providing a direct measurement of the desired polarimetric figure of merit. The first approach is very original with respect to standard polarimetric imaging techniques, as it relies on a specific dualfrequency dual-polarization laser illumination of the scene and on the detection of a radiofrequency (RF) beatnote component in the detected optical signal. The second approach represents the first proposal to my best knowledge of direct polarimetric sensing using a computational imaging approach, relying on the disseminating concept of compressed sensing. Lastly, the third technique that will be presented is based on the local analysis of the speckle intensity statistics in the image, and represents the utmost simplification for polarimetric sensing. In the second part, titled Optimal processing of optical signals in non conventional imaging approaches, I will show how some of the well-known tools of information theory can be used in a scientific/engineering approach (i) to characterize the performance of an imaging setup and (ii) to develop efficient signal/image processing algorithms for specific tasks. In this document, the processing tasks considered will range from estimation of parameters, to contrast enhancement of active targets in images and unsupervised model selection in infrared spectroscopic data. After a brief reminder of the main information theoretic tools that have been used in my research (Shannon entropy, Fisher information, Kolmogorov complexity) in Chapter 4, the second part of this manuscript will address optimal extraction of information from non conventional optical signals, in different application domains. In Chapter 5, the issue of imaging through scattering media will be considered, relying on either a polarimetric approach or a fast intensity-modulation technique. I will illustrate how the Fisher information allowed us to define relevant gain criteria in terms of contrast enhancement of a remote target in turbid atmosphere, making it possible to evaluate the potentiality of these imaging setups for navigation assistance through fog. Chapter 6 will present another example of information theory-based processing of unconventional optical signals. In the context of unsupervised wideband infrared spectroscopy, I will show how the principle of Minimum Description Length (MDL), inspired from the algorithmic definition of the information (Solomonoff-Kolmogorov’s information theory), can be implemented to achieve unsupervised model selection, and how it can outperform standard penalized regression techniques for detection and estimation of atmospheric gas concentrations. Lastly, a general conclusion along with scientific perspectives for future work will be provided in the last part of this manuscript.