In this paper, we describe the plan and advancement of a neural network based image retrieval framework for microscopic images using a reference information base that contains images of more than one information. Such an extraction requires a point by point assessment of retrieval execution of image highlights. This paper presents a survey of crucial parts of content based image retrieval including highlight extraction of color and surface highlights. The proposed neural network based image retrieval framework utilizes a multitier way to deal with arrange and recover microscopic images including their particular subtypes, which are generally hard to separate and characterize. Broad examinations on neural network based image retrieval frameworks show that low-level image highlights can't generally depict elevated level semantic ideas in the clients mind. This framework empowers multi-image inquiry to ensure the semantic consistency among the recovered images. New weighting terms, roused from information retrieval hypothesis, are characterized for multiple-image inquiry and retrieval. The multi-image inquiry calculation with the proposed weighting technique accomplishes about normal order exactness at the main position retrieval, beating the image-level retrieval precision by about ideal rate focuses for different infections separately. Utilizing low level highlights just does exclude human insight. In the event that human mediation is permitted in the image retrieval framework the proficiency supports up.