Here are some of the computer vision problems I worked on through various projects/ class assignments. Tool used: MATLAB
Canny Edge Detector was used to detect sharp edges in an image. The five steps involved in this detection are: Smoothing, Gradient Computation, Non Maximum Suppression, and Hysteresis Thresholding. Smoothing is done by applying a 5x5 Gaussian kernel. Sobel Operators were used to compute gradient. NMS was used to keep sharp edges and remove soft ones.
This problem deals with using the Hough Transformation to identify lines in an image.
In this project a single-scale Lucas-Kanade optical flow algorithm and iterative Coarse to fine optical flow method were implemented. The window sizes and tau values (threshold of eigenvalues) were varied to observe response.
This problem dealt with removing dynamic portions of a scene given a sequence of images. The end result was expected to be a static background image, removing all dynamic components (cars moving, people walking etc).
For a Yale face database (5760 images of 10 faces) three techniques were explored for face detection: Naive k Nearest Neighbor and Eigenfaces and fischerfaces. Idea is to represent the input data is by finding a subspace which represents most of the data variance. This can be obtained with the use of Principal Components Analysis (PCA). When applied to face images, PCA yields a set of eigenfaces. These eigenfaces are the eigenvectors associated to the largest eigenvalues of the covariance matrix of the training data. Fischerfaces are basis obtained from LDA. Error rates were observed varying the number of eigenfaces/fischer faces used in reconstruction
