There is always a need to increase the overall relevance of results in Web search systems. Most existing web search systems are query-driven and give the least preferences to the users’ needs. Specifically, mining images from the Web are a highly cumbersome task as there are so many homonyms and canonically synonymous terms. An ideal Web image recommendation system must understand the needs of the user. A system that facilitates modeling of homonymous and synonymous ontologies that understands the users’ need for images is proposed. A Hybrid Semantic Algorithm that computes the semantic similarity using APMI is proposed. The system also classifies the ontologies using SVM and facilitates a homonym lookup directory for classifying the semantically related homonymous ontologies. The users’ intentions are dynamically captured by presenting images based on the initial OntoPath and recording the user click. Strategic expansion of OntoPath based on the user’s choice increases the recommendation relevance. An overall accuracy of 95.09% is achieved by the proposed system.