The most common approach to processing text which originates as a scanned document image is format conversion, in which procedures such as page segmentation and character recognition are used to convert the scanned text into a structured symbolic description which can be manipulated by a conventional text editor. While this approach is attractive in many respects, there are situations in which complete recognition and format conversion is either unnecessary or very difficult to achieve with sufficient accuracy. This paper presents several applications illustrating an alternative approach to scanned text processing in which document processing operations are performed on image elements extracted from the scanned document image. The central and novel insight is that many document processing operations may be implemented directly by geometrical operations on image blobs, without explicit knowledge of the symbolic character labels (that is, without automatic character recognition). The applications are implemented as part of image EMACS, an editor for binary document images, and include editing multilingual documents, reformatting text to a new column width, differential comparison of two versions of a document, and preprocessing an image prior to character recognition.