As　big　data　rapidly　advances,　the　importance　of　cross-modal　retrieval　systems　that　protect　user　privacy　becomes　increasingly　significant.　Current　common　secure　cross-modal　retrieval　methods　primarily　achieve　privacy　protection　by　designing　feasible　encryption　schemes　and　constructing　secure　public　mapping　subspaces.　However,　these　schemes　often　require　a　large　amount　of　plaintext　data　for　training　by　third　parties,　which　can　easily　lead　to　the　leakage　of　data　owners＇　privacy.　To　address　these　security　issues,　this　paper　proposes　a　secure　cross-modal　retrieval　model　based　on　inner　product　function　encryption　(SCMR-IPE),　which　enables　similarity　measurement　between　encrypted　multimodal　data　while　effectively　safeguarding　data　privacy.　Additionally,　it　introduces　a　self-organizing　mapping　network　and　performs　secure　unsupervised　clustering　of　high-dimensional　feature　vectors,　thereby　significantly　reducing　computational　overhead.　Finally,　experimental　results　demonstrate　that　the　proposal　not　only　ensures　user　privacy　but　also　provides　efficient　and　reliable　retrieval　services,　thereby　exhibiting　high　practical　application　value.　　©　2024　IEEE.