Object detection is one of the important tasks in computer vision that has wide range automation like from autonomous driving to medical imaging. While traditional methods like GraphCut (Boykov et al. 2001) have performed well in the low-level task of accurately delineating object boundaries, they often struggle to capture high-level semantic features associated with an image segment. For example, deep learning approaches with Convolutional Neural Networks (CNNs) are great in feature extraction at the cost of potentially good segmentation precision and style transfer methods offer some elegance but lack scalability. The present paper proposes to exploit the attributes of each approach and combines GraphCut with deep learning in a composite object detection model. The proposed model first uses a Convolutional Neural Network for feature extraction from input image. CNNs are known to detect intricate patterns as well as characteristics in images by the means of stacks, which might include layers such as Convolutional, Pooling and Activation functions. These are very important for recognizing the different objects like edges, shapes, and fabrics of an image. After the feature extraction, a Region Proposal Network (RPN) is used to create bounding boxes of prospective objects. The task of the RPN is to scan over the feature maps of different scales output by CNN, and very efficiently see which regions are “likely” containing an object. These bounding boxes are first-guesses of the location of any object, which gives a region to focus on and hence reduces computational overhead leading into more efficient models. The feature maps and the bounding boxes created are then passed to a GraphCut segmentation module GraphCut is an effective segmentation method that employs graph theory to split the input image into foreground and background. GraphCut utilizes not only the feature maps from the CNN, but also spatial information to improve object boundaries around predetermined regions. It finally makes use of Proper boundary keeping in mind the major problem associated with deep learning-based techniques as to How accurately objects has been segmented. It also involved a few post-processing and refinement of the segmented regions with an improved accuracy. This step may include morphological operations for noise reduction, smoothing boundaries and correcting fortuitous introduced artifacts by segmentation. These improvements are key to make the boundaries of final object accurate and consistent. During the final object detection step, refined segments are examined and a yield of detections is generated. If an object detection model is being used, It needs to locate the objects in the image as well. A combination of CNN’s top-down feature extraction and the higher resolution segmentation in GraphCut yields substantial object detection improvements for our model. Beyond these basics are many benefits of this hybrid approach. This model improves on accuracy by combining the semantic information given by CNNs with boundary delineation detail of GraphCut. This combination provides a state-of-the-art model that is robust to variations in object appearance and the background clutter, making it useful for many real-world applications. Furthermore, it is flexible and can be tuned by training the CNN model with various object detection tasks and changing GraphCut parameters. To conclude this paper, a new object detection model that fuses the advantages of traditional segmentation and deep learning is presented. Therefore, the integration of CNNs and GraphCut creates a complete approach for enhancing object detection both in terms of accuracy as well efficiency. This combination should be able to give a good impact on areas that need accurate and robust object detection, like self-driving cars predictions, medical imaging etc.