New Adaptive Filter For The Removal Of Mixed Noise In Images With Fine Detail Preservation. S. SARASWATHI JANAKI * AND D.EBENEZER Department of Electronics and Communication Engineering College of Engineering, Anna University Guindy, Chennai-600 025 INDIA Abstract:- In this paper, we developed an algorithm to remove additive mixed noise in images with the preservation of fine details and edges. The noise characteristics may vary in the same application from one image to another. In these environments, nonlinear general filters will not perform well and adaptive non-linear filters are best suited. The algorithm based on local statistics such as signal variance and noise Variance is considered. It depends on minimum mean square estimation of the corrupted signal. The signal variance and noise variance are calculated through moving signal window and moving noise window respectively. This offers optimal adaptive filtering in the homogeneous regions as well as in the edges. The fine detail preservation has been obtained by a combining algorithm. The performance of the filters in the presence of different types of noise are evaluated and compared with general mean, bi- directional median filters and median filters. The image enhancement factor has been calculated as the performance-measuring factor. A remotely sensed image has been considered to carryout the subjective and objective analysis of the proposed filter . Key-words:- Adaptive max /median, bidirectional median filter, Image enhancement factor, Mean, Median, Nonlinear filter. * Saraswathi Janaki ,Senior Lecturer , Department of ECE, Crescent Engineering College, currently deputed to pursue research in Anna University, Guindy, Chennai,, India. 1 Introduction The nonlinear filters are optimized for signals having specific statistical charecteristics. Images are non-stationery processes and their statistics vary in the various regions. The noise characteristics may also vary from one application to another. Therefore adaptive is natural choice in this case. The filtering algorithm is based on local statistics with and without threshold. The scheme is based on Minimum mean square estimation of the information- bearing signal corrupted by additive noise. Small windows are used over the signal to obtain the signal mean and standard deviation. Similarly small windows are used over the corrupted signal to obtain the noise mean and standard deviation. Bidirectional Median filters are proposed instead of general median filters for the estimation of signal statistics corrupted by noise. The objective of the paper is to eliminate mixed noise (i.e.,) impulse noise and white Gaussian noise or uniform and impulse noise or mixed of all the three. The adaptive nonlinear filters work well in the removal of noise as well as in preserving the fine details and edges. 2 Algorithm for the Removal of Noise (without threshold)