In my denoising wavelet filter (first I apply a row filter,then down sampling, then  applied an colomn filter to the resulted image, then down sampling , the appliead the  thresholding. My doubt is that how  determine the  threshold( can we calculate from the filtered image ?)

DWT is an advanced form of low/high pass filters. Under the assumption that signal and noise have different levels of spectra, you are tring to find a spectrum level of noises. I am not sure what is actually going on in your algorithm, but i can guess that estimation of spectrum of noises is more likely meaningful when you apply it on the original image rather than the filtered image.
But my opinion is purely based on speculation, I am not sure the correct answer for your case at this point.

My algorithm is given below

Noisy image ->high pass row filter ->{ high pass filtered output -------->down sampling------>high pass colom filter .....(1)

                                                     --> { low pass filtered output--------->down sampling -------->high pass colom filter .......(2)

....(1) { high pass colom filter output---->down sampling---->thresholding

....(1) {low pass colom filter output ---->down sampling---->thresholding

....(2) { high pass colom filter output ------->down sampling ----->thresholding

....(2) {low pass colom filter output------>down sampling -------->thresholding

low pass row/colom filter output =image (before the row/colomn high pass filter applied) - row/colomn  high pass filter output  .

In this filter there are four output cmponents

For reconstructing the orginal image the reverse operation is done.

My doubt is  how determine the   threshold value ? (if threshold value is high some image data is losed, if low some noise is there.)

 

Thanks for the post. It is really an interesting topic.

It's been a long while to think about DWT. I just looked at my code to recall that I implemented hard thresholding, soft thresholding, wiener thresholding, etc. However, your question is how to measure the threshold, which has not an easy answer due to data dependency of thresholding process. My code shows that I estimated the threshold value in wavelet space (i.e. in processed image) as average + (standard variation) * multiplication factor, where average and standard deviation are measured from wavelet coefficients. In other words, I used order statistics on wavelet space. Let me know if there is any problem, I will appreciate it.

 

I used the formula average +standard varation * multiplication factor as my threshold value. Then one problam is that my image infomation also lost( that means the threshold value is high.). So I change the equation as    (average +standard variation)*multipliacation factor.Then the reconstruction of image almost same as orginal. Any wrong in use this formula ?