I finally worked around the extreme darkness of the images that I’m getting from Aibo’s camera. Also, I applied some noise reduction. But the results are still quite poor. The image I’m getting is a 412×318.. although the camera specs say 416×320. In any case, it’s a very cheap CMOS camera, it already has a bad pixel, plus some static pattern noise !
Colors are further washed out by the JPEG compression, which is very happy to destroy pixels with low brightness. It’s a sad situation.
Right now I have two ideas in mind. First is to see if there is something I can do about getting JPEG images with better quality (the functionality is in the API, why it doesn’t work could be my fault). The second idea is a bit more strange. It resembles to white balance. With white balance one sets the brightness and color level of the whole image based on some selected pure white source. My idea is to extend the concept to each pixel. Basically, I would have Aibo looking at some fairly uniformly colored image. A capture of that image would show how far each pixel is off from the source. From that I could create a mask of pixels to be applied to reverse the effect of the static noise of the camera, and the bluish halo around the corners of the image, generated by the plastic in front of the camera.
From the programming point of view, this is easy. At the practical level though. I would need to set-up a properly lit reference target. That probably requires bright white and well diffused lighting.
ummmmm !!

Posted by Davide Pasca in Uncategorized