This　study　proposed　a　method　to　solve　the　problems　existing　in　depression　recognition,　which　is　based　on　visual　information,　improved　particle　swarm　optimization　algorithm　(PSO)　and　support　vector　machine　(SVM).　The　PSO　algorithm　easily　falls　into　local　optimums;　therefore,　to　solve　the　problem,　we　proposed　an　adaptive　mutation　PSO　algorithm　(AMPSO)　to　balance　the　capability　of　local　exploitation　and　global　exploration,　thus　creating　a　classification　model　with　optimal　parameters.　First,　we　used　no-iterative　algorithms　the　kernel　ridge　regression　and　random　forest　to　classify　the　depression　and　normal.　Then,　we　compared　the　recognition　accuracy　using　different　PSO　algorithms　and　found　the　visual　information　accuracy　of　the　AMPSO　algorithm　for　the　SVM　classifier　to　be　the　highest.　Our　research　is　of　an　important　reference　value　for　the　establishment　of　methods　for　depression　recognition　with　clinical　applications.