Genetic diseases are characterized by the presence of genetic variations. These variations can be described in the form of copy number. Microrray-based Comparative Genomic Hybridization is a high-resolution technique used to measure copy number variations. However, the observed copy numbers are corrupted by noise, making variations breakpoints hard to detect. In this paper, we provide a framework for the analysis of copy number. The first part of the framework uses an extended version of nonlinear diffusion filter as pre-processing technique to denoise the observed data base. The extension accounts for the nonuniform physical distance between probes. The second part uses estimates the relative frequency of local and global genomic variations across multiple samples to identify statistically and biologically significant variations. For evaluation, we provide copy number variations results using simulated and real data samples. We also validate the predicted copy number variation segments of copy number gain and copy number loss using the experimental molecular tests quantitative polymerase chain reaction and show that our proposed approach is superior to popular commercial software.