This paper proposes a novel hybrid feature set for image retrieval by combining classical feature engineering techniques with deep convolutional neural networks. The goal is to improve the efficiency of handcrafted features in image retrieval systems. The authors analyze the efficiency of feature generation layers in common deep networks such as residual, classical sequential convolution, and bottleneck. They remove the classification layers in most popular deep networks and use the output of feature generation layers at different depths, converted to feature vectors via a flattened layer, to connect with handcrafted features. Various popular handcrafted features in the spatial domain, such as color, texture, and wavelet in frequency, are used to generate handcrafted features. The proposed method combines residual block feature maps with handcrafted features, which is shown to improve the performance of image retrieval systems compared to bottleneck and sequential convolution layers. The method is evaluated on benchmark datasets, Corel-1 and Corel-5k, achieving accuracy of 96.68% and 94.56%, respectively. The results show that combining residual block features with handcrafted features improves the performance of image retrieval systems. The method is compared with state-of-the-art machine learning and deep learning techniques, showing promising results in terms of precision and recall rates. The main contributions of this paper are: (1) comparing the capability of feature generating layers in deep neural networks to increase the efficiency of image retrieval systems; (2) improving the performance of handcrafted features such as color and texture in image retrieval systems by adding the output of residual blocks in the feature fusion phase; (3) reducing the semantic and content gap in handcrafted-based image retrieval systems, with residual block outputs being more efficient than popular convolutional and bottleneck layers; and (4) providing a comprehensive overview of image retrieval methods using handcrafted and deep learning-based features. The paper aims to propose an efficient image retrieval method by combining handcrafted features with the output of popular feature generating layers in deep networks.This paper proposes a novel hybrid feature set for image retrieval by combining classical feature engineering techniques with deep convolutional neural networks. The goal is to improve the efficiency of handcrafted features in image retrieval systems. The authors analyze the efficiency of feature generation layers in common deep networks such as residual, classical sequential convolution, and bottleneck. They remove the classification layers in most popular deep networks and use the output of feature generation layers at different depths, converted to feature vectors via a flattened layer, to connect with handcrafted features. Various popular handcrafted features in the spatial domain, such as color, texture, and wavelet in frequency, are used to generate handcrafted features. The proposed method combines residual block feature maps with handcrafted features, which is shown to improve the performance of image retrieval systems compared to bottleneck and sequential convolution layers. The method is evaluated on benchmark datasets, Corel-1 and Corel-5k, achieving accuracy of 96.68% and 94.56%, respectively. The results show that combining residual block features with handcrafted features improves the performance of image retrieval systems. The method is compared with state-of-the-art machine learning and deep learning techniques, showing promising results in terms of precision and recall rates. The main contributions of this paper are: (1) comparing the capability of feature generating layers in deep neural networks to increase the efficiency of image retrieval systems; (2) improving the performance of handcrafted features such as color and texture in image retrieval systems by adding the output of residual blocks in the feature fusion phase; (3) reducing the semantic and content gap in handcrafted-based image retrieval systems, with residual block outputs being more efficient than popular convolutional and bottleneck layers; and (4) providing a comprehensive overview of image retrieval methods using handcrafted and deep learning-based features. The paper aims to propose an efficient image retrieval method by combining handcrafted features with the output of popular feature generating layers in deep networks.