KMeans for big data using preconditioning and sparsification, Matlab implementation
Computer Vision - Local Features (HARRIS, MSER, SIFT, PCA-SIFT, GLOH)
This project uses K-Means clustering, a machine learning algorithm for clustering the pixel in an image. Works pretty good in colorblind testing images. It can also be used for determining the quantization level of colors in image and helps to provide a system to minimize the memory required for saving the image.
We propose a superpixel-based fast FCM (SFFCM) for color image segmentation. The proposed algorithm is able to achieve color image segmentation with a very low computational cost, yet achieve a high segmentation precision.
Learn how to read, load and visualize point clouds using MATLAB and pre-process the data by down sampling and de-noising. You will also learn how to apply affine transforms like translation and rotation. Finally, you will learn how to fit point clouds to geometric shapes and how to extract a region of interest from images using point clouds.
Image Classification Based on Bag of Features with SIFT and SURF Descriptors
Interest point detectors and descriptors are at the heart of most computer vision applications. The goal of this project is to gain insights by working on one such application in determining homographies. By building a brief descriptor and matching feature points, the project is able to calculate homogeneous matrix between two pictures taken by rotating the camera around center. Thus, when given several such images, the system can output a panoramic image which combine all the images together.
Content-Based Image Retrieval (CBIR) is a challenging task which retrieves the similar images from the large database. Most of the CBIR system uses the low-level features such as color, texture and shape to extract the features from the images. In Recent years the Interest points are used to extract the most similar images with different view point and different transformations. In this project the SURF is combined with the color feature to improve the retrieval accuracy. SURF is fast and robust interest points detector/descriptor which is used in many computer vision applications. To improve the performance of the system the SURF is combined with Color Moments since SURF works only on gray scale images. The KD-tree with the Best Bin First (BBF) search algorithm is to index and match the similarity between the features of the images. Finally, Voting Scheme algorithm is used to rank and retrieve the matched images from the database.
Harris corner detector is used to find the region of interest. SIFT descriptor is used to generate fingerprint around the interest point. RANSAC algorithm is used to fit the Homography Transform model.
C++ code for "GMS: Grid-based Motion Statistics for Fast, Ultra-robust Feature Correspondence"
Creating Panorama of images after perspective projection, non max suppression, corner detection and stitching of images.
