Pranav Mishra

Video Codec Optimizations on Cortex A8

With the continuous enhancement in the speed and architecture of mobile processors it has become possible to play even high resolution videos on them. However the compilers have lagged behind in taking full utilization of the processing capabilities of the underlying hardware. H.264, known for its high compression ratio and computational complexity, needs performance optimizations for a computationally constraint environment. Some of these techniques are described in this paper.

An efficient red eye reduction technique

Automatic Red Eye Reduction (RER) technique is one of the essential components for imaging device with flash. We present an efficient RER technique that performs red eye detection followed by a verification procedure and an artifact free red eye correction method. The integral projection based method of red eye detection is fast and efficient in handling wide range of variations in pose, scale, and image quality. To reduce the false positives a novel verification method is used. Finally a simple color correction technique followed by a gaussian smoothing is shown to provide a natural looking red eye correction. To reduce the computational complexity, red eye detection is performed on low resolution image and map them efficiently to high resolution image. The proposed method has been compared with state-of-the-art commercial solution and it is found have better RER accuracy. The overall solution is computational cost effective which makes it feasible for use in low power imaging devices.