Search

Local image features have become ubiquitous for a wide range of computer vision tasks. For embedded and low power devices, speed and memory efficiency is of main concern, and therefore, there have been several recent attempts to improve these issues. In this paper, we are concerned with the early components of the object recognition pipeline, namely, feature detection, feature description, and feature tracking. In particular, we propose a novel approach to speed up feature detectors and to inform feature tracking that speeds up the recognition process by using the concept of adaptive sampling. We select two examples of visual algorithms to be modified by adaptive sampling and present comparative results with and without modifications. We show how processing time and memory footprint can benefit by this approach with little impact on overall output quality. We implement our proposed methods on a chipset commonly found on smartphones and we discuss the obtained improvements. [ABSTRACT FROM AUTHOR]

In this paper, we study the rate-distortion (RD) optimization of the H.264 main profile encoding. Specifically, a soft decision quantization (SDQ) algorithm is developed based on the context adaptive binary arithmetic coding (CABAC) method in the H.264 main profile. Given motion prediction and quantization step sizes, the proposed SDQ algorithm is proved to achieve near-optimal SDQ for residual coding in the sense of minimizing the true RD cost when the weak adjacent block dependency utilized in CABAC is ignored for optimization. The SDQ algorithm is then used in conjunction with a general RD optimization framework to jointly design motion prediction and residual coding for H.264 main profile coding given previously coded reference frames. Experiments have been conducted based on the reference encoder JM82 of H.264 main profile. Comparative studies show that the joint design method achieves on average 10% rate reduction at the same PSNR when compared with the RD method in the 11.264 main-profile reference software, with half of the reduction coming from the proposed SDQ algorithm, and 20% rate reduction at the same PSNR when compared with the RD method in the H.264 baseline-profile reference software. [ABSTRACT FROM AUTHOR]