Content-Based Image Retrieval (CBIR) is becoming a significant field of studies owing to the enormous increase in multimedia content in this digital globe today. In most of the CBIR system, features are extracted based on the grayscale image. And the color features are obtained from distinct color channels while using the color image, which lacks the correlation between the distinct channels. In this paper, the concept of a quaternion matrix is used to represent the correlation of color image, and a CBIR system is proposed using Quaternion Discrete Cosine Transform (QDCT) with Bag of visual words (BoVW). First, the stable keypoints are obtained by using a feature detector, i.e., Scale Invariant Feature Transform (SIFT). Further, considering the stable key points as the centre pixel, QDCT is applied in the centre pixel as well as in their surrounding pixel. Subsequently, QDCT coefficients are chosen based on the zigzag scanning order and the feature descriptor is constructed, which is further used to describe the BoVW. The experimental results indicate that the newly proposed retrieval technique produces better results.