A fast video compression method using spatiotemporal segmentation based on Hilbert scanning

We investigate a novel video compression method using the Hilbert curve. This curve is applied to various problems such as image compression, the traveling salesman problem, and database management, because of its locality preservation property. We map the three-dimensional data to one-dimensional data along the Hilbert curve and then apply a lossy compression method using a linear approximation to the one-dimensional data. This method uses simple segmentation of the one-dimensional data and does not require complex computation such as DCT or motion estimation. Experimental results show that our method obtains acceptable quality of reconstructed images comparable to H.263 at low bit rates but is about ten times faster than H.263 (full search motion estimation). It is suitable for video telephones which require real-time encoding of video signals.