During acquisition, transmission, storage and retrieval, quite often digital images become noised and mask the important details of an image and also present themselves with an ugly look. Such images if not processed properly, become useless for subsequent image processing operations like segmentation, classification etc. Hence the quality improvement (through denoising) of such noisy images becomes an important operation in the field of image processing. This paper proposes a two stage detection-estimation based filtering system to suppress random valued impulse noise from digital grey and color images based on neighborhood statistics of pixels of the working window under consideration. An adaptive working window and the adaptive tolerance computed from the pixels belonging to the current filtering window facilitates an appropriate detection of noisy pixels in the first phase, followed by the simple and computationally efficient restoration strategy in the second stage. The Simulation results prove that the proposed scheme exhibits much superior performance in comparison with other state of art image filtering methods in suppressing the random valued impulse noise from the digital grey and color images.