Query expansion is a long standing relevance feedback technique for improving the effectiveness of information retrieval systems. Previous investigations have shown it to be generally effective for electronic text, to give proportionally better improvement for automatic transcriptions of spoken documents, and to be at best of questionable utility for optical character recognized scanned text documents. We introduce two corpus-based methods based on using a string-edit distance measure in context to automatically detect and correct transcription errors. One method operates at query-time and requires no modification of the document index file, and the other at index-time and operates using the standard query-time expansion process. Experimental investigations show these methods to produce improvements in relevance feedback for all three media types, but most significantly mean that relevance feedback can now successfully be applied to scanned text documents.