The creation of forged images in the realm of digital media presents considerable obstacles to the genuineness and integrity of visual content. Deep learning, particularly through the use of convolutional neural networks (CNNs), has become a significant advancement in the field the automated identification of counterfeit images. The paper presents a thorough examination of recent progressions and obstacles in employing deep learning models for the detection of counterfeit images. Initially, an overview is provided on the prevalent categories of counterfeit images, encompassing manipulated images, deepfakes, and artificial images produced by generative adversarial networks (GANs). Moreover, an investigation is carried out into the diverse forms of input data utilized for training deep learning models, encompassing image pixels, metadata, and semantic attributes. The MaxPooling2D layer is chosen for dimension reduction. Furthermore, an analysis is conducted on the crucial role played by extensive 140 K dataset consisting of face images facilitating the training and assessment of deep learning models for counterfeit image detection. An examination is conducted on the assessment metrics and procedures commonly applied for evaluating the efficacy of counterfeit image detection algorithms. The proposed model outperforms other existing models in terms of accuracy to detect fake images by reducing the spatial dimensions of the feature maps. Classification accuracy of 99.78% is achieved and in turn reduces the computational complexity of the model which helps prevent overfitting by focusing on the most important features.