Sony Corporation

Toru Kurata

Ming-Chang Liu

Yoshihiro Murakami

Yutaka Yoneda

Mark Allen Robertson

A method of improving accuracy and reliability of motion estimation is described herein. In one aspect, a 2D neighborhood of phase correlation peak is approximated with an outer-product of two 1D vectors to eliminate the sub-pixel error. In another aspect, estimation of reliability is improved. In yet another aspect, two-pass phase correlation is implemented to eliminate sub-pel motion bias.

Method to improve accuracy and reliability of motion estimated with phase correlation

A method and system to improve the performance of phase correlation motion estimation for low-bit-precision implementation are described herein. Phase correlation uses the Fast Fourier Transform (FFT) with operations with infinite-precision constants. Since physical implementations use finite-precision arithmetic, there is some loss in precision relative to the ideal infinite-precision case. In low-complexity implementations, it is desirable to use as few bits as possible, and if the precision is too low, the performance of traditional phase correlation suffers. A pre-processing technique is applied to the data prior to taking the FFT, which minimizes the negative effects of finite precision in the FFT and allows high quality results from phase correlation. The pre-processing step is a content-dependent contrast adjustment that maps the range of the input images' pixel values to the range of input values for the FFT. There is no post-processing required after the FFT to compensate for the pre-processing step.

Method to increase the accuracy of phase correlation motion estimation in low-bit-precision circumstances

Phase correlation is an established method for computing motion which relies on the ability to find peaks in a computed phase correlation surface. Two methods to improve the ability to detect peaks in the phase correlation surface are described herein. The first method applies a theoretically-derived and spatially-varying gain to the phase correlation surface. The gain compensates for peaks whose amplitudes have been decreased due to windowing effects; such effects are unavoidable in phase correlation. The second method uses concepts from matched filters to improve detection of peaks whose amplitudes are diminished due to a spreading of the peak energy into surrounding positions in the phase correlation surface. Peak detection filters allow such low-amplitude peaks to be properly detected. It is possible to use only the first method, or only the second method or both methods combined.

Growing community of inventors

Ageo, Japan

Toru Kurata