Abstract: Based on the characteristics that the farmlands and ponds have significant corners in high-resolution image, the extraction of farmlands and ponds is achieved accurately in this paper. During the corner extraction procedure, according to the sensitivity of the corner direction change, traditional algorithm was improved with bidirectional analysis technology and more robust bidirectional TK (Tomasi-Kanade) corner detector was proposed. COVPEX (Corner validation based on corner property extraction) was used to validate result of corner detection. Authors found not only the pseudo-corners which were almost useless for image analysis but also the corner-clusters were presented in the validation result. The pseudo-corners reduce rationality of the results, and the corner-clusters destroy the uniqueness of corner. To solve the problems, COVPEX algorithm was improved by associating with multi-scale analysis technology, multi-scale COVPEX algorithm and the “removal corner-cluster” operation were proposed. The results showed that the accuracy and rationality of corner extraction were significantly improved.