CODEWITHSUNDEEP
matlabresizemachine-learningcomputer-visioncrop
buggedUp
buggedUp

Reputation: 103

Does the sequence of image resizing and cropping matter?

I have some images of different sizes. I also have the cropped images and the dimensions, where they were cropped. Now, I want new cropped images with size x*y. It can have a little more or little less area than the previously cropped image. Should I resize the original images and then crop to avoid artifacts on boundaries? If yes, what size should I resize the original image and then what size to crop? The aspect ratio should be preserved.

Upvotes: 0

Views: 643

Answers (2)

olgis
olgis

Reputation: 96

Why do you want to re-size or crop them!? If you do this to increase a number of examples. A good way to get additional positive examples without affecting quality is to mirror your training images across the vertical axis (so that a left-upper corner will become a right-upper one). A mirrored example is different from the original but equally likely to occur, so mirroring is a good way to generate new data.

As well which algorithm do you use!? If you use LDA, LDA requires fixed length features, as do most optimization and machine learning methods.

Upvotes: 0

Dennis Jaheruddin
Dennis Jaheruddin

Reputation: 21563

In general fewer steps will provide a better result. Especially increasing size will not work out very well in many cases.

However, this is just in theory and the difference in quality might not be visible at all. The best way to find out how it will turn out for your case is just to try it as @slayton mentioned.

Upvotes: 1

Related Questions