The classical image processing is based on the evaluation of data delivered by a vision sensor systemas images. The captured light information is extracted sequentially from each photosensitive element(pixel) of the matrix with a fixed frequency called frame rate. These data, once stored, form a matrixof data that is entirely updated at the acquisition of each new image. Therefore, for high resolutionimagers, the data flow is huge. Moreover, the conventional systems do not take into account the factthat the stored data have changed or not compared to the previously acquired image. Indeed, there is ahigh probability that this information is not changed. Therefore, this leads, depending on the "activity"of the filmed scene, to a high level of temporal redundancies. Similarly, the usual scanning methodsdo not take into account that the read pixel has or not the same value of his neighbor pixel read oncebefore. This adds to the temporal redundancies, spatial redundancies rate that depends on the spatialfrequency spectrum of the scene. In this thesis, we have developed several solutions that aim to controlthe output data flow from the imager trying to reduce both spatial and temporal pixels redundancies. Aconstraint of simplicity and "Smartness" of the developed readout techniques makes the differencebetween what we present and what has been published in the literature. Indeed, the works presented inthe literature suggest several solutions to this problem, but in general, these solutions require largesacrifices in terms of pixel area, since they implement complex electronic functions in situ.The operating principles, the emulation in MATLAB, the electrical design and simulations and theexperimental results of the proposed techniques are explained in detail in this manuscript
