This paper concerns a pyramidal representation of images. The objective of this study is to represent an image with as small a number of data words as possible in a memory without distorting the feature extraction facility inherent in pyramidal representations. First, a novel scheme for simplifying pyramidal representations is presented. Simplification is essentially based on the reduction of the number of informative elements observed in a window space during the hierarchical process of image representation. Following a brief explanation of the window pattern simplification process is a description of an overall procedure for simplifying complex binary images. Finally, for images which are not necessarily binary, procedure for rapid access to and selective use of particular areas and local control of detailed fineness are discussed in conjunction with the architectural considerations for implementing the system.