A universal noise filter based on neighborhood feature matching

Image de-noising is an important step in remote sensing image restoration. The main challenges of de-noising algorithm design preserve as more texture details as possible while removing image noise. Due to poor imaging environment and remote transmission of image data, remote sensing images typically contain a strong mixture of Gaussian and impulse noise. The existing de-noising algorithm which can remove the mixture of Gaussian noise and impulse noise is rare. The basic concept of non-local means is followed. By introducing the concept of neighborhood similarity evaluation and impulse noise detector, the universal noise filter based on neighborhood feature matching is proposed. The experimental results show that the filter can effectively remove mixed Gaussian and impulse noise, and maintain good image texture, and transplant easily to DSP/FPGA multi-processor platform.