Method of estimating motion in interlaced video

Abstract

A method of estimating motion in interlaced video involves, firstly, a frame search ( ), where a search is conducted for the frame structure using a sub-sampled block matching metric (e.g. sub-sampled SAD). The locations to be searched are either fixed or dynamically determined based on the minimum frame SAD (ie best frame block match). Next, pixel errors are calculated ( ) for all pixels included in the sub-sampled block matching metric. Each pixel error is first identified as belonging to one of four field patterns (e.g. even-even, even-odd, odd-even and odd-odd). For each location, pixel errors can be classified into two field patterns, either even-even and odd-odd or even-odd and odd-even ( ). The pixel errors belonging to the same field pattern are added together ( ) to obtain field error values (eg field SAD values). The individual field error (or field SAD) values are used to determine ( ) the field Motion Vectors. The location of the lowest field SAD value is taken as the position of best match. All pixel errors are then summed together ( ) to obtain the frame error and the location of the lowest frame error (or frame SAD) is taken as the position of best match for the frame. The frame and field Motion Vectors can be refined by using a full block matching metric within the small search window ( ).