find a invoice slip on a package

Given :- A image of a package

Goal :- To localize the invoice slip on the package

Tried corner detection , thresholding , morphological operations but couldn't reach the desired output

Input :-

Desired Output :-

Scripts (Not giving desired output) :-

Unrecognized function or variable 'filename'.

I = imread("image");
I=im2gray(I);
    level = graythresh(I);
    BW = I>(level*max(I(:)));
    BW=imfill(BW,'holes');
    BW=bwareaopen(BW,1000);
    se=strel('rectangle',[21 21]);
    BW=imdilate(BW,se);
%     BW = bwconvhull(BW);
    CC = bwconncomp(BW);
    
    numOfPixels = cellfun(@numel,CC.PixelIdxList);
    [unused,indexOfMax] = max(numOfPixels);
    BW(CC.PixelIdxList{indexOfMax}) = 0;
    img=zeros([size(BW,1) size(BW,2)]);
    img(CC.PixelIdxList{indexOfMax})=1;
    stats = regionprops(img, 'BoundingBox');
    crop_region = stats.BoundingBox; % contains [top_left_x top_left_y width height]
    cropped_img = imcrop(I, crop_region);

Answer 1

[David Willingham]

If traditional image processing doesn't work, you could possible try to build an object detector based on Deep Learning.

The key to building an objector detector is to have enough labeled data to train the model. The Image Labeler app will help here, as you can upload all your images and label them efficiently.

[Yash Verma]

We were thinking of going with YOLO is it a good option ?

[David Willingham]

Yolo is great but you have a few options. Here is a link to the pretrained models in MATLAB you can use as a starting point: https://github.com/matlab-deep-learning/MATLAB-Deep-Learning-Model-Hub#ObjectDetection