Electromagnetic radiation reflectance increases dramatically around 700 nm for vegetation. This increase in reflectance is known as the vegetation red edge. The NDVI (Normalized Difference Vegetation index) is an imaging technique for quantifying red edge contrast for the identification of vegetation. This imaging technique relies on reflectance values for radiation with wavelength equal to 680 nm and 830 nm. The imaging systems required to obtain this precise reflectance data are commonly space-based; limiting the use of this technique due to satellite availability and cost. This thesis presents a robust and inexpensive new terrestrial-based method for identifying the vegetation red edge. This new technique does not rely on precise wavelengths or narrow wavelength bands and instead applies the NDVI to the visible and NIR (near infrared) spectrums in toto. The measurement of vegetation fluorescence has also been explored, as it is indirectly related to the efficiency of photochemistry and heat dissipation and provides a relative method for determining vegetation health. The imaging methods presented in this thesis represent a unique solution for the real time monitoring of vegetation growth and senesces and the determination of qualitative vegetation health. A single, inexpensive system capable of field and greenhouse deployment has been developed. This system allows for the early detection of variations in plant growth and status, which will aid production of high quality horticultural crops. / Master of Science

No / The application of computer vision in face processing remains an important research field. The aim of this chapter is to provide an up-to-date review of research efforts of computer vision scientist in facial image processing, especially in the areas of entertainment industry, surveillance, and other human computer interaction applications. To be more specific, this chapter reviews and demonstrates the techniques of visible facial analysis, regardless of specific application areas. First, the chapter makes a thorough survey and comparison of face detection techniques. It provides some demonstrations on the effect of computer vision algorithms and colour segmentation on face images. Then, it reviews the facial expression recognition from the psychological aspect (Facial Action Coding System, FACS) and from the computer animation aspect (MPEG-4 Standard). The chapter also discusses two popular existing facial feature detection techniques: Gabor feature based boosted classifiers and Active Appearance Models, and demonstrate the performance on our in-house dataset. Finally, the chapter concludes with the future challenges and future research direction of facial image processing.

The task of determining the types of objects present in the scene, or object recognition is one of the fundamental problems of computer vision. Applications include, medical imaging, security, and multi-media database search. For example, before attempting to detect suspicious behavior, an automated surveillance system would have to determine the classes of objects that are attempting to interact. This task is adversely affected by poor quality video or images. For my thesis I addressed the problem of differentiating between pedestrians and vehicles in both Infra Red and Electro Optical videos. The problem was made quite difficult by the targets: small size, poor quality of the video as well as the precision of the moving target indicator algorithm. However combining the inverse wavelet transform (IDWI) for feature extraction and the Support Vector Machine (SVM) for actual classification provided results superior to other features, and machine learning techniques.

As more universities begin to offer distance education classes, advances in current methods of delivering classroom information must be introduced. At present, universities use two different methods for distance education. The first is text-based web pages, which due to bandwidth restrictions are generally unable to display complex multimedia information. The second method is to videotape lectures and distribute the tapes to distant sites. While this does a reasonable job of simulating a classroom, the cost associated with producing and distributing the videos and the delay involved in distribution makes this system unattractive. This thesis presents a method for compressing the classroom video to a smaller size so that the lecture can be rebroadcast over the Internet without losing classroom information. In addition, methods will be demonstrated for automatically extracting various types of information from a videotaped lecture; this will result in a more interactive lecture than a simple videotape would provide.

For detecting difficult objects, such as hands, an algorithm is presented that uses tokens and a grammar. Tokens are found by employing a new scale space edge detector that finds scale invariant features at object boundaries. First, the scale space is constructed. Then edges at each scale are found and the scale space is flattened into a single edge image. To detect a hand pattern, a grammar is defined using curve tokens for finger tips and wedges, and line tokens for finger sides. Curve tokens are found by superimposing a curve model on the scale space edge image and scoring its fit. Line tokens are found by using a modified Burns line finder. A hand pattern is identified by parsing these tokens using a graph based algorithm. On a database of 200 images of finger tips and wedges, finger tip curves are detected 85% of the time, and wedge curves are detected 70% of the time. On a database of 287 images of open hands against cluttered backgrounds, hands are correctly identified 70% of the time.

The environment in which robotic fruit harv_e~ters work presents many challenges. Developers of a system that aims at performing in this environment need to take into account its inherent variability and devise their system to be robust enough to perform acceptably in all cases. This is no easy task, and a lot of work remains to be completed before the perfect fruit-harvesting robot is designed. This thesis attempts to take a step towards this goal by outlining an algorithm that would allow a vision-controlled robotic harvester to detect and accurately locate oranges in the canopy of an orange tree_ Most of the other researchers that have addressed this issue did not tackle the problem of accurately locating the oranges in an image of an orange tree, only differentiating the orange regions from the rest of the picture. In a previous work at UCF by Joakim Eriksson a system was developed that accurately locates the single, non occluded oranges in a color image of an orange tree. In this thesis, this initial work will be improved upon to give the system the capability of finding occluded oranges, and oranges inside a cluster increasing the overall detection accuracy at locating oranges. Accurately pinpointing the location of the orange would allow a robotic harvester to cut the stem of the orange by either scanning the top of the orange with a laser or by directing a robotic arm towards the stem to manually cut it. Future work needs to address the collection of the oranges after the robot has harvested them.

The problem of tracking moving objects in a video sequence is a well known and well researched problem in Computer Vision. Tracking moving objects is a basic tool which allows the development of solutions to complex problems such as target acquisition, automatic surveillance, action recognition, etc. Tracking problems and solutions generally deal with video that has relatively good frame-rates, i.e. from 15 to 30 frames per second, and the objects in motion do not exhibit huge jumps. However, if the video frame rate is low or, more precisely, the objects in motion move large distances from frame to frame, current tracking methods will perform very poorly This thesis proposes a method of tracking that will allow for large spatial discontinuities in object motion and still be able to track successfully. It demonstrates the feasibility of tracking in these sequences. Results are given from application of the proposed method to video sequences taken at 2 frames per second.

The quality of an input sketch can affect performance of the computational algorithms. However, the quality of a sketch is not often considered when working with sketch tasks and automated sketch quality prediction has not been previously studied. This thesis presents quality prediction on the "Sketchy" dataset. The method presented here predicts a quality label rather than a zero to one quality metric. This thesis predicts an understandable label rather than a computer-generated quality metric with no human input. Previous tasks like sketch classification have used a transformer architecture to leverage the vector format of sketches. The architecture used in sketch classification was called Sketchformer. The Sketchformer was adopted and trained to predict quality labels of hand-drawn sketches. This Sketchformer architecture achieves 66% accuracy when predicting the 5-labels. The same transformer achieves up to 97% accuracy in a different experiment when combining the different labels into good versus bad (2-label) experiments. The sketchformer significantly outperforms the SVM baseline. The results of the experiments show that the transformer embedding space facilitates separation of 'good' sketch quality from 'bad' sketch quality with high accuracy. / Master of Science / If pictures are worth 1000 words, then sketches are worth a few hundred words. Sketches are easy to create using a pen and tablet. Objects in the sketches can be drawn many ways, depending on the talent of the creator and pose of the object. The quality of the sketches vary pretty drastically. When using sketches in computer vision tasks, the quality of a sketch can affect the performance of the computational algorithm. However, the quality of a sketch is not often considered when working with other sketch tasks. One common sketch task is called Sketch-Based Image Retrieval (SBIR). The input of this task is the sketch of an object/subject, and the model returns a matching image of the same object/subject. If the quality of the input sketch is bad, the output of this model will be poor. This thesis predicts the quality of sketches. The dataset used is called the "Sketchy" dataset, this dataset was originally used to study SBIR. However, the creators of the dataset provided quality labels for the sketches. This allows for quality prediction on this dataset, which has not previously been completed. There are 5 different labels assigned to sketches. One of the experiments completed for this thesis was predicting 1 of the 5 labels for each sketch. The other experiments for this thesis create good and bad labels by combining the 5 labels. The Sketchformer architecture created by Ribeiro et al. is used to run the experiments. The Sketchformer achieves 66% on the 5-label experiment and up to 97% on the good and bad (2-label) experiment. This transformer outperforms a Support Vector Machine baseline on this quality labels. The results of the experiments show that the transformer applied to this dataset is a valuable contribution by surpassing the baseline on multiple tasks. Additionally, accuracy values from these experiments are similar to values found in the corresponding image quality prediction task.

Language provides us with a powerful tool to articulate and express ourselves! Understanding and harnessing the expressions of natural language can open the doors to a vast array of creative applications. In this work we explore one such application - natural language based image editing. We propose a novel framework to go from free-form natural language commands to performing fine-grained image edits. Recent progress in the field of deep learning has motivated solving most tasks using end-to-end deep convolutional frameworks. Such methods have shown to be very successful even achieving super-human performance in some cases. Although such progress has shown significant promise for the future we believe there is still progress to be made before their effective application to a task like fine-grained image editing. We approach the problem by dissecting the inputs (image and language query) and focusing on understanding the language input utilizing traditional natural language processing (NLP) techniques. We start by parsing the input query to identify the entities, attributes and relationships and generate a command entity representation. We define our own high-level image manipulation language that serves as an intermediate programming language connecting natural language requests that represent a creative intent over an image into the lower-level operations needed to execute them. The semantic command entity representations are mapped into this high- level language to carry out the intended execution. / Master of Science / Image editing is a very challenging task that requires a specific skill set. Hence, Going from natural language to directly performing image edits thereby automating the entire procedure is a challenging problem as well as a potential application that could benefit widespread users. There are multiple stages involved in such a process starting with understanding the intent of a command provided in natural language, identifying the editing tasks represented by it and the different objects and properties of the image the command intends to act upon and finally performing the intended edit(s). There has been significant progress in the field of natural language processing as well as computer vision in recent years. On the natural language front computers are now able to accurately parse sentences, analyze large amounts of text, classify sentiments and emotions and much more. Similarly on the computer vision side computers can accurately identify objects, localize them and even generate real life like images from random noise pixels. In this work, we propose a novel framework that enables us to go from natural language commands to performing image edits. Our approach starts by parsing the language input, identifying the entities and relations in the image from the language followed by mapping it into a set of sequential executable commands in an intermediate programming language that we define to execute the edit.

Computational vision and image analysis is a multidisciplinary scientific field that aims to make computers "see" in a way that is comparable to human perception. It is currently one of the most challenging research areas in artificial intelligence. In this regard, the extraction of information from the vast amount of visual data that are available today as well as the exploitation of the resulting information space becomes one of the greatest challenges in our days. To address such a challenge, this thesis describes a very general computational framework that can be used for performing efficient inference and learning for visual perception based on very rich and powerful models.