Image processing has a wide range of applications and is applicable to almost all domains. With the availability of cognitive services, it can be augmented to build far-reaching solutions. Some areas where it is applicable are medicine, video analysis, pattern recognition, robot vision, remote sensing, and a host of other areas. And since image processing is a compute-intensive process, algorithms and methodology used can be critical.
One use case of image processing is in processing application forms, in which various documents like driver’s licenses, passports, etc. are submitted along with application forms, which are scanned and stored. These images need to be classified as per their classes for further action. The application forms will need to be read and identified. The use case can further be extended, for example, to answering natural language questions from application form documents, identifying objects, making robots understand what they see, etc.
Convolutional neural networks using Keras libraries in Python can be used to classify images. Convolutional neural networks are low-compute, pre-trained models and will take less time and resources compared to other image-classification techniques. A specific machine-learning model is required to be built and trained on a set of data before using image classification. Then, given a set of images, the machine-learning model will list classify images accordingly.
The code pattern titled “Image classification using convolutional neural networks” demonstrates the use of convolutional neural networks and Keras for image classification. Check out the pattern to understand how to build and run an image classifier using convolutional neural networks and Keras.