Graphical User Interface (GUI) testing has been the focus of mobile app testing. Manual test cases, containing valuable human knowledge about the apps under test, are often coded as scripts to enable automated and repeated execution using test harnesses to reduce the cost. Unfortunately, many test scripts may become broken due to changes made during app updates. Broken scripts are expected to be updated for reuse; however, the maintenance cost can be high if large numbers of test scripts require manual repair. We propose an approach named METER to repairing obsolete test scripts automatically when mobile apps are updated. METER novelly leverages computer vision techniques to infer GUI changes between two versions from screenshots and uses the GUI changes to guide the repair of test scripts. In experiments conducted on 18 Android apps, METER was able to repair 78.3% broken test scripts automatically.