FiveFocal has developed image signal processing algorithms for three principal classes of end use – task specific computer vision, human observation, and test and characterization. Our team has additionally developed these algorithms while taking into consideration the customer’s hardware implementation constraints.
Algorithms for Images and Video that will be Observed by Humans
In general, algorithms that fall into the human observation category have the goal of creating a pretty picture, or clearly displaying the pertinent information in a scene. Within the human observation class of algorithms, FiveFocal has experience developing the following:
- Demosaicing – Linear and non-linear demosiacing methods for Bayer and alternate color filter patterns that balance image resolution and aliasing.
- Noise Reduction – Non-linear noise reduction and signal extraction algorithms that are spatial, spectral and signal dependent, resulting in the key balance of image quality and required computation.
- High Dynamic Range (HDR) – Image synthesis and tone mapping from both spatially and temporally multiplexed sensors as well as from HDR sensors.
- Gamma – Used to encode linear luminance or color information for transmission, storage or display
- Sharpening/Deconvolution – Sharpening/deconvolution methods for both traditional and computational imaging systems with the goals of naturally presenting imagery while minimizing spatial artifacts.
- Sharpness metrics for autofocus – Metrics for image based (passive) autofocus for use in single and multi zone architectures in addition to supporting static and servo modes.
- Lens Shading – Correcting for the light falloff from both imaging lens and microlens effects
- Pixel Defect Correction – Locating and correcting defects in the pixel array
- AWB and Color Correction – Producing the most accurate color possible under varying lighting conditions
- AEC/AGC – Determining the optimal gain and exposure setting for best quality images.
- Compression – Prior to storage or transmission the image or video data is typically compressed to minimize the size of the data while maintaining adequate image quality.
- Multispectral (from Visible to LWIR) – These general classes of algorithms have been developed in the visible, near IR, SWIR, MWIR and LWIR spectrums.
In addition to the basic building blocks, FiveFocal has worked on integrating the various components into optimized ISPs to maximize performance while minimizing the demands and requirements on the available hardware.
Task Specific Algorithms
FiveFocal has broad experience with a large class of task specific algorithms including:
- Barcode reading – Optimizing and preconditioning input data from traditional and computational imaging systems for effective use with state-of-the-art barcode recognition software.
- Iris Recognition – Optimizing and preconditioning input data from traditional and computational imaging systems for effective use with state-of-the-art iris recognition software.
- Simultaneous Location and Mapping (SLAM) and Parallel Tracking and Mapping (PTAM) – Strong understanding of current SLAM and PTAM algorithms and their applications, tradeoffs, and optimal requirements of the vision system
- Feature Extraction and Identification – Search and identification of features within a general scene.
- Tracking – Tracking of an object of interest within a video stream.
Algorithms for Characterization and Test
- High Speed MTF Calculations from Test Charts – Robust slanted edge MTF algorithm development in addition to approximations to this calculation for hardware and rapid throughput applications.
- Distortion Calculation from Test Charts – Calculation of lens distortion.
- Color Fidelity Measurement – Calculation of color fidelity based off of test charts.
If you are interested in discussing your tool, signal processing or consulting needs, CONTACT US.