Image search via Radon Barcode projections Barcode
The generation of a barcode to represent images has many practical applications, both in image representation as well as search queries based on an image. A Radon barcode is a type of barcode designed to represent the contents of an image in a simplified yet comprehensive form. This barcode can then be used to conduct searches based on an image, finding images that have been found to produce similar barcodes. The use of a barcode to conduct the search over other commonly used alternatives, such as those that look for commonly occurring shapes in the content of an image, can be advantageous due to the speed at which barcodes can be searched.
- Abbas Rizvi
- MohammadHamza Asif
- Shahzabe Mahmood
- Jad Eletry
The Radon Barcode application functions through three main methods.
The generate barcode method reads in an image and creates a barcode to represent the data within the image. This is done through applying several transformations and converting each row into a binary number.
After barcodes are generated for each image, the barcodes are then put through a search function against other barcodes in the dataset and verifies if the search resulted in the same number being found. The successes are tallied and used to create a metric regarding the accuracy.
For this dataset the accuracy is 65%, however this can be improved at the cost of run time.
Radon barcode technology can be used in many different places to provide fast and efficient querying of large amounts of data. While not as accurate as modern AI models, it can still be tuned to perform better and will be able to better scale with further training and refinement.
python 3.10
- Clone and enter the repository
$ git clone https://github.com/abbas-rizvi/RadonBarcode.git
$ cd RadonBarcode
- Extract MNIST dataset
Extract MNIST_DS.zip file into directory
- Install Requirements Install dependencies from requirements.txt using pip
$ pip install -r requirements.txt
- Run radon.py
$ python radon.py