High quality image acquisition system based on deep-learning

Nowadays, cameras are widely embedded into many various devices like phones, wearable devices, robots, and cars and they are used in a large variety of applications providing efficient tools for improving the perception capabilities of these devices. In order to achieve this, they are introducing deep-learning approaches to improve quality performance of traditional computer vision tasks. However, these approaches are well known to perform very well under optimal conditions, i.e., input images/videos without noise, high quality, etc, but such conditions are not obviously always met when these cameras are operating under real-word environmental conditions (what we call environmental noise), including rain, dust, fog-haze, fire-smog, light reflections, limited illumination conditions, etc.; under well-known hardware camera limitations (i.e., sensor noise, non-linearity, white balance, image artefacts); or under security bridges (e.g., adversarial attack). In this proposal, we will use the term of adversarial conditions to identify any of these three types of noise. 

The aim of this internship is the development of a novel algorithm to guarantee reliable quality of acquired data, obtained independently and irrespective from the type of adversarial conditions under which the camera is operating. This data can be further stored, manipulated, and transmitted without reducing the high quality of the acquired image/video. This will guarantee that computer vision task-related applications, using the data acquired by the camera, will be capable to maintain their recognition capabilities even in the presence of any adversarial conditions where the data are taken.  

• Programming language Python,
• knowledge of deep-learning tools keras, 
• tensorflow 

Good

The objective is to use deep-learning approaches to improve the image quality acquition of camera independently from which adversial conditions it is operating. 

Expected deliverables:prototype sw