CS6420 -- Presentation Schedule

Nora Alosily
Video Segmentation via Temporal Pattern Classification
April 17, 2017
Video shot boundary detection or the process of breaking up a video into units that have coherent contents is important for further video analysis. This method exploits low-level feature vectors to build intermediate features, using pairwise similarity. The result is passed on to a supervised KNN classifier that will be used subsequently to detect shot boundaries in a given video.

John Brandenburg
Illumination-Aware Age Progression
April 12, 2017
Age progression techniques have applications beyond entertainment and can be used to help solve missing children cases. We will present an approach that takes a single photograph of a child and automatically produces a series of age-progressed outputs between 1 and 80 years of age accounting for pose, expression and illumination. Using a large dataset of individuals we will produce averages based on clusters of age ranges, then find the differences in shape and texture between the input photograph. Applying these differences will yield the age progressed photo.

Lina Deng
Car License Plate Detection
May 01, 2017
Car license plate detection is an emerging area of research due to various applications such as prevention of crime, electronic toll system, intelligent traffic control system etc. The method is after converting the color input image into grayscale, an adaptive thresholding is used to obtain the binary image. Then the unwanted lines are removed through an unwanted-line elimination algorithm. Finally to detect the license plate, vertical edges are detected by Sobel operator.

Tzu-An Song
Local Binary Pattern Edge-Mapped Descriptor Using MDM Interest Points for Face Recognition
May 03, 2017
Present local methods such as local binary pattern (LBP), local derivative pattern (LDP), and scale-invariant feature transform (SIFT) perform better than holistic methods; however, their high complexity results in some limitations for applications such as mobile devices. In addition, SIFT-based schemes are sensitive to illumination variation. We propose an LBP Edge-mapped descriptor by using Maxima of Gradient Magnitude (MGM) points. It is a robust, simple and fast descriptor. LBP Edge-mapped descriptor is a string of binary codes which record surrounding information of illumination and edges of MGM. It can illustrate facial contours completely and have low computational complexity simultaneously.

Sandhya Vaidyanathan
Enhancement of Fingerprint Images
April 26, 2017
Fingerprints are widely used for biometric identification due to their consistency and uniqueness during a person's lifetime. The enhancement of fingerprint images helps us to obtain better results during classification of fingerprints. This enhancement method includes but not limited is techniques like normalization, orientation estimation using Sobel operator, Gabor filters, binarization, and thinning.

Ameet Vishwakarma
Image Super-Resolution Using Deep Convolutional Networks
April 19, 2017
This project aims at a method to recover high resolution image as output from a single low resolution image as an input. This method uses deep learning method for image super resolution (SR), it directly learns an end-to-end mapping between the low- and high-resolution images with pre/post processing beyond the optimization. It features a deep convolutional neural network (CNN) of a lightweight structure, yet demostrates state-of-the-art restoration quality, and achieves fast speed for practical usage over traditional models. This could help in reducing the data traffic due to images over the network to a great extent.

Sherd White
Automating Image Morphing Using Structural Similarity on a Halfway Domain
April 24, 2017
The goal is to create and automate image morphing using mapping to line up image elements such as color and geometry transitions between two images. The map is created by analyzing for corresponding similarity of image segments. Optimization occurs through the use of interpolation and smoothing based on a selection of points selected by the user to identify and aid in the image transition. The calculations can be done in parallel on a GPU for increased speed.

Da Zhao
SIFT Flow: Dense Correspondence across Different Scenes
April 10, 2017
While image registration has been studied in different areas of computer vision, aligning images depicting different scenes remains a challenging problem, closer to recognition than to image matching. Analogous to optical flow, where an image is aligned to its temporally adjacent frame, we propose SIFT flow, a method to align an image to its neighbors in a large image collection consisting of a variety of scenes. For a query image, histogram intersection on a bag-of-visual-words representation is used to find the set of nearest neighbors in the database. The SIFT flow algorithm then consists of matching densely sampled SIFT features between the two images, while preserving spatial discontinuities. The use of SIFT features allows robust matching across different scene/object appearances and the discontinuity-preserving spatial model allows matching of objects located at different parts of the scene. Experiments show that the proposed approach is able to robustly align complicated scenes with large spatial distortions. We collect a large database of videos and apply the SIFT flow algorithm to two applications: (i) motion field prediction from a single static image and (ii) motion synthesis via transfer of moving objects.