Indian Coin Denomination Dataset

Inspiration:

The primary motivation behind this dataset is to foster the development of innovative computer vision solutions for automated currency recognition, which is a critical need in many industries, including retail, banking, and vending machine technologies. The Indian Coin Denomination Dataset provides a well-organized and detailed foundation for training models to accurately identify and classify coins based on their denominations.

Dataset Content: The dataset comprises high-resolution images of Indian coins of the following denominations: One (1) Rupee, Two (2) Rupees, Five (5) Rupees, Ten (10) Rupees, and Twenty (20) Rupees. Only one side of each coin—the tail side, which clearly displays the denomination—is captured in this dataset. The images were collected using a mobile device, with the coins placed on a white A4-sized sheet of paper to provide a clean and consistent background for improved visual clarity.

Sampling and Variations:

A total of 30 distinct coins were sampled, representing the full spectrum of denominations included in the dataset. The breakdown of the coin samples is as follows:

• 7 One Rupee coins
• 4 Two Rupee coins
• 9 Five Rupee coins
• 6 Ten Rupee coins
• 4 Twenty Rupee coins

Each coin was photographed from 5 different viewing angles and under 3 different lighting conditions, including variations in brightness and shadow intensity. Every combination of viewing angle and lighting condition was captured twice, resulting in a total of 900 images. The dataset is well-balanced across the denominations, with each class represented as follows:

• One Rupee: 210 images
• Two Rupees: 120 images
• Five Rupees: 270 images
• Ten Rupees: 180 images
• Twenty Rupees: 120 images

Data Collection Process:

The dataset was created using a controlled environment to ensure uniformity across the images. Coins were placed on a flat surface (white A4 paper), and photos were taken using a mobile phone camera to simulate a common data collection scenario. Images have a resolution of 3000 x 4000 pixels, providing high levels of detail that facilitate accurate classification and detection tasks.