Stephen Rhein Email & Phone Number

Phoenixville, Pa, Us

San Jose, Ca, Us

Contributing to the research and development of high content image analysis products for the quantitative analysis of protein and cell biology.

San Francisco, California, Us

Develop and optimize computer vision algorithms for Augmented Reality applications.

Lake Mary, Fl, Us

Designed and developed an embedded computer vision system through cross functional new product development with firmware engineers. Failures of existing computer vision systems were documented and solutions consisting of algorithmic and optimization improvements were proposed, implemented, and benchmarked. The algorithms consisted of customized pattern recognition, feature extraction, and matching as well as an automated testing framework. Optimizations were made through GPU parallelization using OpenCL, CPU vectorization using NEON intrinsics, CPU parallelization using Intel Thread Building Blocks, and algorithmic changes. Development also relied heavily on Java, OpenCV, C++, and Python libraries.

Newark, De, Us

Developed frameworks and algorithms for the quantitative analysis of fungi in gigapixel scale plant microscopy 3D image stacks. Extended NumPy with support for a reference counted, shared memory array using POSIX interprocess communication for Python multiprocessing: http://git.io/xQo8. Using the ctypes module, NumPy arrays were passed through a C library interface to C++ code implementing various image processing algorithms. Tasks were distributed using intra-node multi-process pools where message passing of serialized objects was minimized by the use of our shared memory library. The package scikit-learn was used to classify infections, and ImageJ and Java were used for additional image analysis functionality.

Newark, De, Us

Developed algorithms for the analysis of large datasets of polar sea-ice images. Used Intel TBB for synchronized, data-flow concurrency in image acquisition and analysis: http://git.io/xQoQ. A proof-of-concept framework was developed for a four camera system (infrared and visible light cameras) to be triggered from software and processed concurrently for real-time rendering and persistent storage. An extension was made to support hardware triggered cameras where frames were received asynchronously but joined using Intel TBB flow graph nodes. Various metadata from cameras and images were serialized using Boost for persistent storage. The metadata was managed in Python through SQLAlchemy.

Newark, De, Us

Instructed undergraduate students in Object Oriented programming using Java. Created lecture slides as well as exercises, homework assignments and exams.

Newark, De, Us

Accelerated the detection of hidden data in large images using massively parallel graphics processors. As a precursor to the application of machine learning techniques, feature extraction throughput was improved 80x using CUDA for NVIDIA GPUs and the speed of histogram computation was doubled through custom GPU vectorization. The computation of features made use of the GPU memory hierarchy by managing the caching of subproblem solutions in on-chip memory. For computing histogram bin offsets, multiple pixels were packed into a single register and operated upon in a SIMD fashion to limit register pressure and instruction count. Inline PTX assembly was then used to optimize the reduction of results into a global solution.

Coached computer science students in graduate/undergraduate courses and participated in various aspects of class administration.

Contributed to a $1-million effort to combat cyber-terrorism using digital video steganalysis. The FFMPEG video processing library was modified to hide data in MPEG compressed videos. By extending an MPEG-2 codec written in C, a 5-tuple parameter vector guided the motion compensated prediction of inter-frame movement to embed a hidden signal. Videos modified in this way were then trained and tested using Support Vector Learning Machines to gauge the extent to which the presence of hidden data was undetectable. A 1-week computer security bootcamp was provided to underprivileged students to educate them on these and other relevant computer security issues.

Milton, Pennsylvania, Us

Lead development of a business application for job quoting of energy production equipment. Modeled the job quoting workflow relationships and developed a web interface using CakePHP. An entity relationship diagram constructed using Dia was automatically converted to an SQL database schema. This schema was then introspected/reflected by CakePHP to construct project scaffolding for customization.

Master Of Science (M.S.), Software Engineering

Education record

Full Stack Web Development Certification, Computer Software Engineering

Master Of Science (M.S.), Computer Science

Bachelor Of Science (B.S.), Computer Science