Don, I have been "thresholding" which I believe is the typical name for the process of converting an rgb image to a b&w image using the ideas I have read about in Cliff Reiter's book entitled, "Fractals, Visualization and J." In that book he defines the following terms and I use the "threshold" 110 below to *train* my playing card system.
avg =: +/ % # wb_pal =: 110>avg"1 read_image jpath'~....' To *use* my playing card system I have found that a similar threshold value is required but depends on the lighting system in the room. Marshall's comments referred to "intensities" which I take to mean are the values returned by the avg"1 verb above, before that result is compared to 110 or whatever the threshold limiting value is. In this thread/conversation I was seeking a better way to determine that threshold limiting value. (I had missed his initial reference to intensities, which slowed my understanding of his approach. I have not confirmed with Marshall that avg"1 produces "intensities" but that is what I have assumed.) Do you have some other measure of lighting that can be read from an image? Something different or better than the avg"1 I am using? Also, are you suggesting that this global assessment of the threshold value is unwise and that it needs to be done in patches around the image? If so, could you elaborate, please? I have begun to use Marshall's idea but have been rather frustrated because although it seems to work to determine the value of the cards' suits, it is often not so good for determining the pips' values. I am still experimenting, but tentatively I am subtracting about 10 from the suggested threshold value from the diamonds. My system is frustrating because I cannot create a physical situation where I can control the lighting and the positioning of each playing card so that the cards can be readily passed in front of the webcam one-at-a-time and accurately recognized. I am limiting myself to almost-free or already purchased objects I find around the house. This is all just for fun, not for profit. My current webcam is an apple iSight which sits on a pedestal and is connected to the cpu via a cable. It is very limited in its deployment/positioning flexibility but at least I can trigger a snapshot from my desktop computer with the Chrome browser. I own a clear plastic dual card deck tray but cannot think of any way to point the camera directly down to see the top card in the tray, for example. I guess I need to build a rig of some sort to hold the webcam above the deck tray. I think the better thing to do is use an ipad or iphone and its camera, but as of this date, there is no way to trigger the ipad camera from the web browser (I am using jhs, btw). Building a rig for an ipad/iphone would seem easier. I don't do android, but that may be an alternative, except I don't believe android browsers can trigger the system camera either. Well, that's undoubtedly much more that you (all) wanted to know. On Wed, Jun 28, 2017 at 12:29 PM, Don Guinn <[email protected]> wrote: > Are you losing information by applying the threshold to each point without > regard to the points around it? If most or all the points around it are > large values then it is more likely that the threshold should be lower than > if the point is surrounded by lower values. Also, maybe look at the overall > brightness of the image to help set the threshold. > > -- (B=) ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
