augmented reality
In this project, I implemented algorithm for camera calibration using OpenCV in C++ and applied Augmented reality by rendering virtual object on image plane using transformation matrix in real time. I also developed Face Augmentation by detecting Face using Haar Cascade classifier and projecting sunglasses on face.
object detection
For real-time 2D object detection, I programmed image segmentation, filtering and feature extraction with OpenCV in C++ for object detection. It achieved accuracy of 92% in detecting object in real time with modeling using NN and KNN classifier.
recognition using cnn
Recognition using CNN (Convolutional Neural Network) is a popular approach for image recognition tasks. CNNs are a type of deep learning neural network that is inspired by the structure of the human visual system. They are designed to learn hierarchical representations of visual features in an image. I modeled CNN to recognize the digits by training the network on MNIST dataset. It The CNN layers helped to extract features from the input image and the extracted feature vector is then fed into classification layers and finally predicting the label of the input image. Upon training the network for MNIST data, transfer learning was done to classify the Greek letters. Finally, the hyperparameters tuning was done using Random search and optimized the network performance for MNIST Fashion data.
content-based image retrieval
I implemented Content-based Image Retrieval (CBIR) techniques that can search through the Image database and provides the best match to the queried image. It was based on the query-by-example method in which the user provides a target image and the objective was to display the most similar based on the contents and features of the image. I build robust feature vectors by building color and texture histograms with Gabor filters. The features were compared using distance metrics such as Euclidean distance, Manhattan distance, and Histogram intersection.