Go to DBLP homepageICDRF: Indian Coin Denomination Recognition Framework
Abstract: Coins are commonly used in everyday transactions at supermarkets, retail shops, metro stations, and other locations. However, recognizing Indian coin denominations poses a challenge for visually impaired persons due to the lack of a robust coin denomination recognition system. To address this issue, we introduce the Indian Coin Denomination Recognition Network (ICDRNet), a deep convolutional neural network (CNN) tailored for the visual recognition of Indian coin denominations. ICDRNet combines densely connected convolutional layers with depthwise separable dense blocks, incorporating the Convolutional Block Attention Module (CBAM) and a novel Dilation Enabled Inverse Bottleneck (DEIB) block. The integration of CBAM enhances the model’s ability to highlight critical features across different coins by prioritizing essential spatial and channel information, thereby improving recognition accuracy. The DEIB block employs dilated convolutions to efficiently capture detailed features across various scales, focusing on subtle distinctions among coin denominations with minimal computational overhead. Furthermore, we present the Indian Metal Currency Dataset, a comprehensive collection of images representing Indian coins under diverse lighting, background, and environmental conditions, facilitating a robust evaluation of coin recognition performance in realistic scenarios. Experimental results show that the proposed method shows improved performance in identifying coin denominations compared to prior methods on the proposed Indian Metal Currency Dataset (IMCD), Indian Coin Currency Dataset (ICCD), Indian Coin Denomination Dataset (ICDD) & Comprehensive Image Dataset of Contemporary Indian Coins (CIDCIC) datasets. ICDRNet achieves an impressive performance with F1-score of 97.79%, 98.38%, 96.7% and 99.7% on IMCD, ICCD, ICDD & CIDCIC datasets, respectively.
External IDs:dblp:journals/access/KanrooKRG25
Loading