How to Extract a certain Shape from Image

Question

Max 2013-2-18

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/63907-how-to-extract-a-certain-shape-from-image

评论： Darshan Drviruz 2017-10-21

I would like to distinguish between "background" pixels and the "orange plastic bag" pixels in order to manipulate those pixels (count, hist, etc).

I can't use indx = I(I ~= 1) because there are some white pixel inside the orange plastic bag.

Moreover, this bag is aligned but it might be rotate in other images.

Thanks.

Image URL: http://snag.gy/tk7kH.jpg

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Rakshit Kothari 2013-2-18

So what you basically need is some form of object recognition, am i correct? If you just need to extract the bag of chips, image segmentation using Otsu's method is probably what you need.

If you need recognition i suggest you read up on SIFT or SURF.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Walter Roberson 2013-2-18

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/63907-how-to-extract-a-certain-shape-from-image#answer_75464

convert to gray, threshold to select non-white, imfill() 'holes'. The result should be a mask you can use to select from the larger image.

2 个评论
显示无隐藏无

Max 2013-2-18

it is a problematic since I can't use indx = I(I ~= 1) because there are some white pixel inside the orange plastic bag.

Walter Roberson 2013-2-18

That's why you imfill() 'holes', to recover the "inside" part.

请先登录，再进行评论。

Answer 2

Rakshit Kothari 2013-2-18

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/63907-how-to-extract-a-certain-shape-from-image#answer_75480

So what you basically need is some form of object recognition, am i correct? If you just need to extract the bag of chips, image segmentation using Otsu's method is probably what you need.

If you need recognition i suggest you read up on SIFT or SURF.

If its a quick thing that you are looking for, just invert the image, threshold and then use REGIONPROPS with filled holes option.

If you need the shape of the object:

Ratio = Area of Object/Area under bounding box

If Ratio is a reasonably high value, you object is a square or a rectangle. If its lower, its probably a triangle or circle.