Information retrieval can be visualized as the extraction of the desired information from the flooded resources that spread over World Wide Web. Image retrievals are the fundamental and critical problem that arises in the retrieval activities. In this regard, it is considered to be a challenging task which requires utmost care. Diverse characteristics of data such as noisy, heterogeneity impose a great barrier over image retrieval applications. This article aims to come up with a state of art approach for overcoming these problems by clubbing together widely recognized deep architecture along with natural language processing. This novel design methodology utilizes the latent query features, deep belief network, Restricted Boltzmann Machine for learning tasks. This collaborative work can be used to reduce the epoch in the learning periods whereas the rest of the methods fail to achieve the constraints.