arablau's profile - activity

Hi Everyone,

I am working on a project right now where I am trying to track features on product photography.

I have a ton of pictures of a purse for example and I want to tag the buckle on the front and have it tagged in all the other images as well.

Right now, I am using SIFT to find the nearest keypoint to where I click and then find that keypoint in other images.

The issue is that sometimes there are no keypoints returned by SIFT around a feature I want to track (like a purse strap or zipper or something).

What other methods are there to doing something like this?

The only two I can think of right now are using keypoints and using the known position of the camera and camera matrices to determine the location of that product feature in other picture.

Any thoughts?

Thank you! Calling initModule_nonfree() worked!

Hi everyone,

I was wondering what the difference is between cv::FeatureDetector and specific classes like cv::SIFT.

If I create an instance of cv::SIFT as follows, my code works:

cv::SIFT sift;

// Detect key points and extract descriptors
std::cout << "Detecting key points" << std::endl;
sift(front, cv::Mat(), frontKeyPoints, frontDescriptors);

but if I try to use the cv::FeatureDetector::create my program crashes!

// Create SIFT feature detector    
siftDetector = cv::FeatureDetector::create("SIFT");

// Detect key points
std::cout << "Detecting key points" << std::endl;
siftDetector->detect(front, frontKeyPoints);

Why does the FeatureDetector crash? Is one better to use than the other?

Thanks

Yes, I want to remove the peach background and make all of those pixels true white. I am playing around with eroding and dilating and filtering but don't know if there is a better way to get just the object.

Hi Everyone,

I'm working on a project where I am photographing products like this purse below.

I want to remove the background and have been trying a few methods but wanted to know what the community thinks is best practice.

I've been playing with different amounts of blurring, thresholding, dilating and eroding but I am not sure what combination of these measures will yield the best results.

The best I can do so far is this:

What are your thoughts?

Thank you all!