m24758 – Stanford University proposes a context based location coding scheme which uses sum of local neighborhood features. Performance can be further improved by taking feature selection measures into account. It suggests also an adaptive RANSAC thresholding depending on quantization block size.
m24782 – Peking University proposes a pruning based location coding scheme that introduces an extra layer of interest gradients to allocate more bits for fine location histogram quantization. It suggests interest area based location coding by providing more budget for the interest points inside interest region subject to the constraints of very low bit rate.
m24802 – Huawei proposes a new histogram map encoding method adopting circular matrix scan, adaptive skip mode and low memory context based entropy coding
The evaluation of performance shown indicates that all proposals perform well and on a very similar level relative to each other. The techniques employed are complementary. m24802 uses circular scan, and the efficiency of is hardware implementation should be evaluated. The group agreed that these proposals should be merged and presented at the next meeting, including full cross verifications.
