In working on enhancing the gray level image using NN, where the input layer depends on the images fetures & statistics and the output layer depends on the enhanced features.
My question is how to reconstruct the image from these fetures? is there a software tool? or method to do that? or any researcher can suggest a better or different way to specify input & output layer
I appreciate your cooperation & help
I reviewed different reconstruction algorithms and their attributes in the following publication:
Conference Paper A REVIEW OF RECENT RANGE IMAGE RECONSTRUCTION ALGORITHMS
Can you give more details, please? What are your features? What kind of statistics do you apply?
Thank you Dr. Volker,
The image features ( average gray level, entropy, variance, standard deviation, skew, intensity, ...etc) insert as input neurons of NN, the output layer will be the enhanced features
Dear Nedhal
As I remember from image processing steps that enhancement proceeds feature extraction.
Feature extraction can be seen as a special kind of data reduction of which the
goal is to find a subset of informative variables based on image data. Since image
data are by nature very high dimensional, feature extraction is often a necessary step
for segmentation to be successful. Feature extraction is also a means for simplifying
segmentation problem.
In this case I would calculate co-occurrence matrix with means, variance, standard deviation, entropy...etc. Or I can calculate principal component.
In your case to create a link I would extract features from the image and based on these features I can find a common relationship that tie these things together. I would consider in your case that I have a weighted objective function with each feature has a weight > 0 and less or equal to 1 ( the weight sum for all feature =1). I assign a weight to each feature according to their influence on the enhancement process and I will try to optimize this objective function.
I can think about an approach that provide vector of multi dimension data of the extracted features as an input to the ANN process such as SOMs. to obtain optimal weights.
Dear Nedhal,
you apply different statistical features which you try to optimize. Therefore, the question arises: What are your optinisation criteria? But anyway, your problem is non-linear and not elementary to invert. One solution is, as M. Awad has descibed, to formulate an optimisation problem which you solve. However, in general it is possible to find a optimisation solution by a Kernel-NN approach. Let me ask ask a simple question. Why do you try to solve an image quality optimisation task with the help of an NN? We have many more straightforward methods in the image processing community available.
Best regards,
Volker LOHWEG
What I understand from your question is that you want to enhance the quality of gray level images based on their statistical features. Your enhancement algorithm therefore, will be optimized using these features which you provide as input to a NN. The NN will give you the optimum parameters for the enhancement. Is that true?
Hello Nedhal,
"reconstructing" in general means to apply the inverse of the original "imaging" function. If you are using imaging functions dependent on the content of the original image, there is no way to reconstruct the original as the imaging function cannot be "guessed" from the imaging result.
Reconstruction does only work if the imaging function is known and the inverse of this function is calculable. ( There are some tricks "tweaking" the inverse function to overcome eg. zeroes of the imaging function (resp. poles of the inverse function) but the above applies in general.)
- Badges
- Science topic