Application of SVD based spatial filtering to video sequences
Konstantinos Konstantinides, Gregory S. Yovanof · 2002
Video capture devices, such as CCD cameras, are a significant source of noise in image sequences. Preprocessing of video sequences with spatial filtering techniques usually improves their compressibility. We present a block-based, non-linear filtering algorithm based on the theories of SVD and compression-based filtering. A novel noise estimation algorithm allows us to operate on the input data without any prior knowledge of either the noise or signal characteristics. Experiments with real video sequences and an MPEG codec have shown that SVD based filters preserve edge details and can significantly improve nearly-lossless compression ratios by 15%.