Abstract: Aiming　at 　the 　critical 　issues 　in 　automatic　3Dbuilding　facades 　and　roofs 　extractionaprelimi nary　3Dbuilding　object 　extraction　al gorithm　from　a　pixel 　hei ght　mapderived 　from　the　obli que　air borne 　imager y　through　dense　matchingwas　therefore　desi gned　and 　implemented．Based 　on　the knowledge　of 　building　structure 　and　surface 　shapesthis　method　takes 　full 　advantage　of 　the　hei ght gradient 　feature 　to 　extract 　the 　facadeswhile　at 　the 　same 　time 　uses 　the　maximum　entropy　threshold feature 　to 　extract 　the 　roof 　surfaces．Multi－angle 　di gital 　aerial 　images 　are 　emplo yed 　for 　experimental 　anal ysisthe 　results 　show　that 　3Dbuilding　objects　with　complex 　shape　can　be　detected　and　extracted reliabl y　from　pixel 　hei ght　map