Monday, December 15, 2008

Omnivision Received EDN's Best Of 2008 Award

Yahoo: OmniBSI has been selected by Electronic Design Magazine as the 2008 'Best Image Sensor Technology'. The publication's annual 'Best Electronic Design' awards are based on editor's choice of the most significant designs and innovations introduced throughout the year. OmniVision's BSI technology was selected for its revolutionary approach to digital imaging.

20 comments:

  1. Good but this company is going down the sewer pipe... So, what if BSI is great! They have not shown that it is commercially viable. 1.4um technology itself is in doubt!

    ReplyDelete
  2. As for me, I admire ambitious projects, even if they end in failure. Time will tell if Omnivision is able to commercialize it, but it certainly worth an award.

    > 1.4um technology itself is in doubt!

    I recall the same words were said about 2.2um pixels. And about 3um before that. History is spiraling, right?

    ReplyDelete
  3. I think it is great that Omnivision got an award for being the first to mass produce BSI cmos imagers but it should not be an innovation award. Anyway, the company probably just applied for the award and there is little if no peer review. Still, first to mass produce is worth something. Nice work TSMC, too.
    -EF

    ReplyDelete
  4. Image Sensor, I love your site! I can't believe how comprehensive it is.

    You mind giving me your opinion on the RIM Storm's video camera? It looks like it's EDoF in video yet autofocuses for stills just like OmniVision's TrueFocus promises. OmniVision's HDR 2mp was in the Bold, so does it have a better than average chance of being in the Storm?

    Have you seen any viable alternatives to TrueFocus with both EDoF and autofocus? The only one I knew of was from DxO/MagnaChip, but we both know what happened to them. The other names I know of are Tessera's OptiML Focus, Dblur, and someone else I can't think of right now (they mostly offer software products). I don't think anyone promises both EDoF and an internal autofocus except for DxO, but I haven't seen anything from them for over a year.

    Here's a video for reference. It looks EDoF to me: http://www.youtube.com/watch?v=iJsiPWn2QPc

    Here's the product brief on TrueFocus (internal autofocus engine): http://ovt.com/data/parts/pdf/OV3642_PB_(1.1)_web.pdf

    Good luck, and thanks in advance!

    ReplyDelete
  5. Congratulation to Omnivision AND TSMC for being the first producing BSI technology. Nervertheless I totally disagree with the referenced article, which insists on higher performances of 1.4µm BSI over 1.4µm FSI (Front Side Imager) and even 1.75µm FSI. If this were the case, so why is the picture taken with 1.4µm FSI on Omnivision site at... 700 Lux!!! It's not low light level. Seriously, congratulation again to Omnivision, but still a lot of work, and FSI is not dead yet!

    ReplyDelete
  6. Regarding RIM Storm's camera, it looks to me that it becomes fixed focus in video mode. It's quite hard to tell as Youtube video has low resolution and nothing seems to be sharp there, but my first impression is that it's fixed focus. Probably it uses AF only in stills mode.

    You are right, there are not that many companies working on EDoF. Many big players are trying to develop EDoF internally, thus avoiding licensing fees. Heriot Watt University works on EDoF as well. As for DxO EDoF, it takes time to design a product chip, so their silence is understandable.

    ReplyDelete
  7. Regarding 700 Lux picture: This is about what they or anybody else need to show near perfect picture. Do you expect them to show less than perfect picture to demo their flagship technology?

    ReplyDelete
  8. Interesting. You think the RIM Storm's fixed focus in video? Have you ever seen another smartphone do that, use autofocus for stills and fixed focus for video?

    I always thought everyone used the autofocus for everything like HTC does for its' Touch Pro: http://www.youtube.com/watch?v=i6-w4XFOkzY

    What's the typical focal range for fixed focus? Would fixed focus be able to see a hand right in front of the lens as well as what's going on in the background like the TV?

    Thanks in advance.

    Your site is truly amazing. I can't believe you stay so connected! Please keep it up!

    ReplyDelete
  9. Not trying to be a pest, but doesn't fixed focus need to be stopped down? I don't think the autofocus portion is stopped down. Can a camera stop itself down at will?

    (OmniVision's HDR 2mp is in the Bold, stopped down): http://www.blackberrycool.com/2008/11/blackberry-storm-vs-blackberry-bold-camera/

    ReplyDelete
  10. That is the huge difference between OV & ST...
    So.. for higher resolution... OV needs more and more investment...

    ReplyDelete
  11. Re RIM Storm AF: You are right, all AF modules I know use AF in preview. I just tried to explain Storm's behavior using Occam Razor principle.

    With low resolution, like QVGA or QCIF, a fixed focus lens is certainly able to keep in focus both hand and infinity. With AF module used set as fixed focus in preview, there is a problem with lens positioning, which Storm's module designers could solve, at least in theory.

    In any case, this is my best guess. But you sound like you definitely know that Storm uses EDoF, do you?

    ReplyDelete
  12. Re OV and ST comparison: Fortunately for Omnivision, it happens to have the money to invest. I wish that ST had it too.

    ReplyDelete
  13. Heh, I wish I knew, but it seems to make sense. I don't know of a single smartphone camera that uses AF for stills and fixed focus for video (not my expertise). Do you know of one? I'd love to see its' video.

    Also, if you throw in the fact that RIM used OmniVision's HDR 2mp in the Bold (according to iSuppli), it might stand to reason that they'd go ahead and make a splash with TrueFocus. After all, they're losing share, and every little edge helps, right? No one else seems to be using TrueFocus.

    Why do you think no one is using the EDoF chips (all two of them?)? All seem capable of EDoF, yet only two have internal autofocus mechanisms, TrueFocus and DxO/MagnaChip's. Could it be that MagnaChip's didn't work, or no one had faith in them considering their closure (which I found out on your site, thank-you very much!), so no one wanted to be stuck with a single supplier, OmniVision?

    -----------------

    OK, now I've got one for you: Nokia was all chummy with Scalado recently indicating that they might use their APIs in their upcoming phones, and Scalado has that technology that can allow for 20 pics per second (or whatever the rate is). Can you use conventional autofocus for that with good and fast quality, or would there be an advantage to using the TrueFocus system, where you can take and store intermediate images quickly and autofocus them later using the ISP?

    Sorry to go nuts, but you seem to be the most knowledgeable sources I can find, anywhere.

    I'll spare asking you if you know anything about the Contrast Transfer Function. ;))

    ReplyDelete
  14. Who is it that is so interested in EDoF? I am sure I can provide some info. There is no EDoF in the 2.0MP RIM device. The downscaling of the 2.0Mp sensor in preview mode provides the good focus. There is no OV EDoF in mass production yet. DXO/MC chip is obsolete and will not be used in any camera.

    ReplyDelete
  15. Would you mind explaining downscaling? Is that different from stopping down the aperture? Do you downscale to extend depth of field? The Bold's camera looks really dark. Does downscaling make images darker than they otherwise would be

    http://www.blackberrycool.com/2008/11/blackberry-storm-vs-blackberry-bold-camera/

    Heh, if OmniVision's TrueFocus was in mass production, we'd all be the last to know.

    How do you know the DxO/MC is obsolete? I know that we're about to move to the 5mp sweet spot, but couldn't it be a substitute for TrueFocus (assuming it works) if it actually was TrueFocus that's in the Storm?

    ReplyDelete
  16. Yes, I'm not aware of any EDoF camera-phone on the market now. However, EDoF phones are nearing and can appear next year.

    Talking about Scalado 20fps speed, I'm not sure that every picture in the burst is focused separately. Most probably there is a single autofocus step at the beginning of the burst, then the focus is kept constant.

    You can read Wikipedia article anout MTF, OTF, CTF and such:
    http://en.wikipedia.org/wiki/Modulation_transfer_function

    ReplyDelete
  17. By downscaling the image from 2.0Mp to QVGA (or other preview mode resolution ) resolution is gained. The spatial frequency is reduced and thus the contrast is increased. This gives sharper, but lower resolution. Downscaling increases the signal, and does not reduce it, so the image can be brighter. There is no change to the aperture stop.

    MC/DXO is a 2.2um sensor. OVT EDoF sensor (to be) is 1.75um. They cannot really substitute each other.

    We will see EDoF camera phone mass-production in January.

    ReplyDelete
  18. Thanks to both of you.

    @image sensor-

    Yeah, I'm only guessing that RIM used the 1/4" 3mp TrueFocus in the Storm since they used OmniVision's 1/5" 2mp HDR in the Bold, combined with the fact that the Storm is AF for stills and either fixed-focus or EDoF for video. Considering that OmniVision is only putting out a TrueFocus 1/4" 5mp and not any regular 1/4" 5mps, it looks at least they are expecting to have EDoF sales.

    Yeah, I've seen that primer on CTFs there and elsewhere. Does that formula make it possible to easily find maximum contrast? Isn't maximum contrast the way that most autofocus systems find optimal focal points? I wonder if OmniVision is thinking about cramming that algorithm into TrueFocus (or creating one), to autofocus instantly. Their patent makes it sound as if it has to compare contrasts of different pictures with different focal lengths to find the best focus.

    @ Anon-

    I'm not following you on downscaling. Downscaling from 2mp to qvga increases resolution? I thought qvga is a lower res, 1/4 of VGA. But it does brighten the signal? So are you saying that the 2mp in the Bold is being stopped down for EDoF but downscaled to get back some of the lost brightness? Is that for both stills and video?

    Ah, I didn't realize that the DxO/MC chip was 1/3". That pretty much made it obsolete from the time they released it, huh? ;)) Even "worse" for OmniVision since they now have absolutely no substitutes. I wonder if the handset manufacturers' apprehension to use TrueFocus is due to that fact (the DxO/MC was the only other EDoF with an internal autofocus).

    I saw the Largan press. Is that who OmniVision's using for their EDoF lenses, or did someone else get some business?

    ---

    Thanks guys!

    ReplyDelete
  19. Most camera-phone AF systems work on maximizing high-frequency contrast component.

    Honestly, I do not know how one can combine AF with EDoF. AF needs a feedback for its iterations, but feedback is blocked by EDoF. Also, the raw image is blurred by EDoF lens, so AF alone can not make it sharp. If you know how to combine these two, please tell me.

    ReplyDelete
  20. I don't know how the DxO/MC proposed to do it (never saw their patent if they even had one), but the TrueFocus patent says that the internal autofocus engine will apply focal points to the intermediate image until it finds the one with the best focus. I'm assuming that means it looks for the one with the largest contrast. The EDoF is just another filter, a straight inverse Fourier of the intermediate image, if I remember correctly.

    That autofocus method is still suboptimal if you ask me. They need to find a way to apply a focus with maximum contrast immediately. Solution in the CTF?

    The patent: http://www.freepatentsonline.com/y2006/0269150.html

    From the patent: "17. A system comprising: a decoder effective to convert a spatially distorted image into an undistorted image; a plurality of filter parameters coupled to the decoder; a passive autofocus (AF) engine, coupled to the decoder, effective to select a processed image as a final image; wherein, in operation, the decoder receives a spatially distorted image and applies the plurality of filter parameters to the spatially distorted image to produce a respective plurality of processed images, and the passive AF engine selects a final image from the respective plurality of processed images."

    Thanks for your help!

    ReplyDelete

All comments are moderated to avoid spam.