Abstarct: Seismic reflection methods are one of the ways to explore the subsurface structures and exploit hydrocarbon reserves. There are always unwanted events in seismic acquisition results. The reliability of seismic mapping is strongly dependent upon the quality of the records. Seismic records are usually affected by various types of noise such as ground rolls, multiples, random noise, reflection and reflected refraction from near surface structures. Random noise resulted from random oscillation during acquiring data is one of the most important and harmful noises that exist in seismic data over entire time and frequency. Random noise attenuation is an important step in seismic data processing affecting the data interpretation. Low signal to noise ratio of the data will have a devastating impact on the interpretation of data, hence necessaries to as denoising methods to enhance the seismic data quality. Different methods with different algorithms have been used for this purpose, each of which have own advantages and disadvantages. Methods developed in the field of image processing can be very useful for this purpose. One of those methods that is effective for reducing random noise, smoothing, although often leads to blurring and loss of fine details. In this thesis, using image processing techniques and filtering baxsed on nonlinear anisotropic diffusion, random noises in seismic data is reduced. These filters are designed baxsed on partial differential equations (PDE). This method with simulation of diffusion process (transport of temperature or mass concentration) attenuates random noise in pictures (seismic section). The diffusion process describes a physical process that balancing concentration changes without creating or destroying mass. Nonlinear diffusion filters regard the (seismic) image as the initial state of a diffusion process which adapts to the evolving image. If we consider the concentration as the intensity of the image then we can use the PDE for designing the diffusion filter types. The results of applying this method on synthetic and real data as well as comparing it with conventional methods such as median filter and F-x deconvolution filter, show that proposed method outperformed the other approaches from the aspects of SNR improvement and increasing the coherency of seismic events.