A parallel processor for motion estimation including: a matrix of elementary processors configured in rows and columns, local connections between the elementary processors for transmitting partial results, and row outputs for outputting a set of best match values (for example, SAD values), one value for each pixel row of a current block of image pixels; and search area delay buffers coupled to each row inputs, for accepting pixels of the search area as input and forming a reference block row. The processor further includes current block delay buffers coupled to each row inputs, for accepting pixels of the current block as input; a sum module coupled to the row outputs for computing a final match value from the row outputs; and a sorting module for sequentially selecting a best match value from the final match values outputted from the sum module, and generating a corresponding motion vector.