"ImageMiner1.4 feature extraction error message"

rapidkhanrapidkhan Member Posts: 1 Contributor I
edited June 2019 in Help
I am using the ImageMiner 1.4.1 with RapidMiner 5.2.0 and trying to use a simple image extraction operatior window_statistics tha takes input
grayscale image plus: expects: ImagePlusGSIOObject. I am able to load up a greyscale image but get the below error with this and other feature extraction operation.

"Process Failed. window dimension must be smaller then window in feature extraction"  ? Does anyone know what it means.. I tried tweakng the window size parameter and even the image size but no help. appreciate any help.



  • Options
    StaryVenaStaryVena Member Posts: 126 Contributor II
    Hi Imran,
    this message means that window dimension defined in Feature Extraction Operator must be greater than window defines in window statistics operator. Otherwise these features will be calculated out of defined bounds and that is forbidden.

  • Options
    hyperplanehyperplane Member Posts: 1 Contributor I
    Imran, what StaryVena says is correct (I think he might be one of the developers - but I am not sure).  But, what he does not mention is that first you need to use a local feature extraction operator and then, nested within that, a feature extraction operator (i.e., your window_statistics operator).  You set window size in BOTH and the outer operator window size must be larger than the nested operator window size.

    Though the image miner extension appears very powerful, I have found it VERY DIFFICULT to work with.  You will discover, if you follow my instructions above, that your process still will not work.  The reason is that the local feature extraction operator outputs an ImagePlusBaseIOObject to the window_statistics operator which, unfortunately, expects a different object, namely an ImagePlusGSIOObject.  This sort of thing happens ALOT with imageminer.  There is only one operator I have found that will accept an ImagePlusBaseIOObject and output an ImagePlusGSIOObject and that is the get_mask operator found in the visualization group.  However, when that operator is inserted into the process, the process still will not work.  In fact, you are met with the same error that you started with!!  (incorrect window size)

    StaryVena did post some examples on MyExperiment.  However, they are of no help.  They apparently use an old version of imageminer and do not work.

    If you figure it out, let me know!

  • Options
    radoneradone RapidMiner Certified Expert, Member Posts: 74 Guru
    Hello hyperplane,
    thank you for your feedback. It is very valuable for us (btw: yes, StaryVena is one of the developers :) The problem is that you are trying to extract local level features from window of dimension mxm, from image of nxn, where m > n. You can scale the image with resize operator. Is it possible for you to provide us an example process and input data where the problem will be presented?

    Please also consider the trainable segmentation (http://www.burgsys.com/image-mining), which is another approach based on local level features and can suit better for your purpose.

    BTW: Rapid-I is currently preparing a book, where we contributed with two chapters about basic and advanced IMMI use-cases. We hope it can be ready on market soon.

Sign In or Register to comment.