Machine Vision Algorithm Advances

TechCrunch: Google Street View team reports that its recent text and street numbers recognition algorithm can be successfully used for solving CAPTCHA puzzles widely used on the web to determine a human versus a spam bot. The algorithm has been presented in a paper at the International Conference on Learning Representations (ICLR).

I have a feeling that this algorithm has leaked outside Google, as starting from about Sept. 2013 the amount of spam comments in this blog has grown dramatically. On same days, per each real comment I get 10 spam ones, in spite of CAPTCHA protection.

Google algorithm recoginizes this with 99.8% accuracy

Google Project Tango Uses Primesense 3D SoC

iFixIt teardown of Google Project Tango 3D camera revealed that it uses Primesense's recent Capri PS1200 SoC and its IR illuminator projects a familiar structured light dot pattern:

Yves Faroudja's Pre-processor Improves Video Compression Efficiency

EETimes: Yves Faroudja presents a layered way to improve the video compression efficiency at NAB. He adds a new pre-processor layer, prior to a standard encoder such as H.264, HEVC, or MPEG 2, and post-processor (after compression decoding). "We take an image and simplify it; and that simplified image goes through the regular [standards-based] compression process," Faroudja explains. "But we never throw away information." This additional information is inserted in what Faroudja calls a "support layer." This compresses signals not used in Faroudja's so-called simplified image. Together with the decompressed simplified image, the support layer helps reconstruct the original image in full resolution and at 35-50% reduced bit rate at the same quality.

Imec's Image Sensor Services

Imec published a brochure about its image sensor services offerings: "Imec offers services ranging from development-on-demand,
over prototyping, to low-volume production." Imec has 200mm and 300mm image sensor fabs with 65nm, 90nm and 130nm processes, including an extensive image sensor toolbox:

• Specific substrate (HR-Silicon, thick or graded dopant epi)
• Pixels (3T, 4T, trench isolation, embedded CCD pixels in CMOS)
• BSI processing
• Hyperspectral filters
• Special ARCs (CMOS compatible)
• Stitching / Butting capabilities for large area imagers
• Micro-bumping, 3D integration with TSVs
• Capability to develop dedicated pixel technologies (e.g. for SPADs).
• Organic imagers (polymer photo diodes & full imagers)

Panasonic Proposes High Resolution Light Field Imaging

Lightfield Forum reviews Panasonic patent application on high resolution light field imaging. The US20140078259 "Light field image capture device and image sensor" by Masao Hiramoto, Yasunori Ishii, and Yusuke Monobe proposes to place microlens behind the photosensitive layer, rather than in front of it like in most other light field imaging approaches. This promotes the Panasonic-Fujifilm organic sensing layer possibilities:

The control layer 1b switches the light reflected from the layer 1c between the 1st and 2nd image capturing sessions:

ON Semi Announces a Family of 4.8um GS Pixel Sensors

Business Wire: ON Semiconductor announces a new PYTHON CMOS image sensor family based on 4.8um global shutter pixels. With resolutions of 300K, 500K and 1.3MP respectively, the PYTHON 300, 500 and 1300 feature in-pixel CDS (ipCDS) - global shutter with CDS in a relatively compact pixel size. The new PYTHON pixel combines a read noise of less than 9 e-, with 7.7 V/lux sensitivity and frame rates as high as 850fps (VGA format). A highly configurable sequencer also allows designers to tailor the sensor operation to the exact needs of the application, including support for fast on the fly updates to the sensor configuration. The sensor's operation is supported across the -40°C to +85°C industrial temperature range.

The new VGA, SVGA and SXGA are the first three sensors in the PYTHON family, with additional higher resolutions planned for release in the near future. All are pin-to-pin compatible with one-another and with the existing VITA1300 image sensor.

"With our new PYTHON image sensors we provide a cost-effective, high performance solution that will address the needs of the growing number of image sensing applications across a range of end markets," said Thad Smith, Director of the Image Sensor business unit at ON Semiconductor. "PYTHON offers an image sensing solution that not only provides appealing levels of speed without compromising image quality, but is also configurable and flexible enough to provide simple integration across multiple resolutions."

Rambus Conditional Reset Patent Application

Rambus PCT application WO/2014/055391 "Conditional-Reset Multi-Bit Read-out Image Sensor" by Thomas Vogelsang, Michael Guidash, Song Xue, and James Harris talks about a version of the old idea to expand DR by resetting pixel each time its signal value crosses a threshold during the integration time: "If the pixel signal exceeds a sampling threshold, the photosensitive element is reset. During an image capture period, digital values associated with pixel signals that exceed a sampling threshold are accumulated into image data."

Powering Image Sensor in Automotive Camera

EETimes Europe publishes an article on how the automotive power circuit for image sensor is built. Nothing unusual there, looks more like an application note. The authors tried to adapt the sensor power to 25mVpp noise coming from a switched regulator. The camera is based on Aptina 1.2MP sensor.

Foveon Quattro Resolution Expanations

Imaging Resource publishes an interview with Foveon's GM Shri Ramaswami explaining resolution enhancement claims of the latest Foveon Quattro sensor. Basically, the reason is that the top high resolution layer is "blue-heavy" rather than blue:

"You can see that the top layer is blue-heavy, but it's not blue. The next layer is green-heavy, but it isn't green. The bottom is red-heavy. None of these is just red, green, or blue -- that really allows you to do something very interesting. So this is fundamentally a pretty smart way to keep your information, but at the same time, reduce the whole load on the system, because these things are not pure colors. It may sound counterintuitive, but it actually allows you to separate out very cleanly all the detail information -- as we call it -- from the top layer, and understand where the color detail comes from. In other words, it allows us to actually get back what was apparently lost." Regarding the possibility to enhance resolution with a pure RGB sensitivity, Shri responds "It would be impossible then, without that correlation, to do this."

Sony Too Applies for GS Pixel with Internal Storage

Sony patent application US20140084138 "Solid-state image sensor, method for the same, and electronic device" by Takashi Machida is quite similar to few other global shutter pixel proposals. I wonder if this patent will be ever granted: