With the enormous growth in the usage and availability of image capturing devices, there exists a proportional arise in storing and searching devices. Such systems have wide range of applications especially in e-commerce, medical image processing, etc. E-commerce is one of the rising fields in which content based retrieval systems (CBIR) plays a vital role. A large amount of memory space is required to store the image features if they are directly used. The redundant information in feature descriptor is the major limitation to reduce searching and retrieval time. So, feature descriptor dimension reduction became an essential part of retrieval system. Once the feature descriptor is extracted the classification technique decides the relevant images for the given query image. At this stage, deep learning concept is necessary to improve the classification accuracy of the system. A trade-off between the retrieval time and precision is required in the present scenario. In this paper, we emphasize the necessity of deep learning in image retrieval and provide an attempt towards the optimum solution to improvise the efficiency of retrieval. The linear discriminant analysis is used for feature space reduction and the SGD-SVM is used for indexing. The system performance is analyzed with different types of datasets such as Flickr-15k and Corel-10k.