Semantic gap, high retrieval efficiency, and speed are important factors for content-based image retrieval system (CBIR). Recent research towards semantic gap reduction to improve the retrieval accuracy of CBIR is shifting towards machine learning methods, relevance feedback, object ontology etc. In this research study, we have put forward the idea that semantic gap can be reduced to improve the performance accuracy of image retrieval through a two-step process. It should be initiated with the identification of the semantic category of the query image in the first step, followed by retrieving of similar images from the identified semantic category in the second step. We have later demonstrated this idea through constructing a global feature vector using wavelet decomposition of color and texture information of the query image and then used feature vector to identify its semantic category. We have trained a stacked classifier consisting of deep neural network and logistic regression as base classifiers for identifying the semantic category of input image. The image retrieval process in the identified semantic category was achieved through gabor filter of the texture information of query image. This proposed algorithm has shown better precision rate of image retrieval than that of other researchers work