UC Riverside

It is common knowledge that conventional camera has a lens and a sensor array as a set. As light passes through the lens, it will be collected and form an image of the photograph subject. Because of the special property of lens, whatever is displayed in front of the sensor will be directly recorded on the sensor array. For lensless camera, the lens is replaced with a binary mask. Unlike the lens can collect light, mask can only project a shadow that human eyes can not recognize on the sensor array and further computational algorithm is required to recover a meaningful image from the shadow. Without the physical constraints of the lens, lensless camera can be extremely flat, light-weight and flexible, which makes it an alternative option to conventional cameras. However, despite these advantages, the quality of images recovered from the lensless cameras is often poor, especially when sensor to mask distance is small or number of sensor pixels is less than the number of scene pixels.

This thesis presents a new method to address the problem of poor reconstruction results by combining coded illumination patterns with the mask-based lensless imaging. Instead of using uniform illumination, the object is illuminated with multiple random generated binary patterns and the camera acquires a sequence of images for different illumination patterns. Apart from solving this problem in a naive way, a low-complexity and recursive algorithm is proposed to avoid storing all the measurements or creating a large system matrix. The results of simulation are presented on standard test images under various extreme conditions and demonstrate that the quality of the image improves significantly with only a small number of illumination patterns.