Tooth uncle tutorial is easy to understand

The test image

Small picture, cropped from the lower right corner of the larger picture

Large image, 1280X720 resolution

There are three types of changes to the big picture

  1. Do not change the big picture, which is divided into two kinds, one is to cut the bottom right corner of the big picture and then compare, one is not cut
  2. Keep the aspect ratio for the larger image and scale to 70%
  3. Instead of maintaining aspect ratio, change width to 1000 and height to 720

Three effects are displayed

1. Do not change the large image and do not cut the large image
The number of optimal matching feature points = 60 Time of finding: 593ms histogram comparison = 0.9420728414825477Copy the code

2. Do not change the large image, but cut the large image
The number of optimal matching feature points = 63 Time of finding: 210ms histogram comparison = 0.9420728414825477Copy the code

3. Scale the large image to 70%
The number of optimal matching feature points = 10 Time of finding: 330ms histogram comparison = 0.8781176823119982Copy the code

4. Change the width to 1000
The number of optimal matching feature points = 4 Time of finding: 543ms histogram comparison = 0.8643715691444815Copy the code

The environment

Phone: Mi 11 Pro

Android version: 11

Autojs version: 9.0.10

Autojs comes with OpencV version: 4.5.1

conclusion

From the above test, SIFT for multi-resolution map, the effect is also ok,

But it’s just ok. In fact, it’s only suitable for some of the image-finding scenarios,

For example, sift doesn’t work with transparent buttons,

The transparent button looks a little bit better with the outline,

In terms of time, the lower the resolution of the large image, the shorter the time consumption,

So when we’re looking for a graph, it’s a good idea to crop out the areas where the smaller graph might exist,

In order to improve the efficiency of map finding,

In addition, I think it is not correct to find the picture only by using the feature points of the picture.

So I added another dimension: the histogram,

Can help determine if the image we find is the one we want,

The other problem is that we have to make sure that we can find the pictures,

And also to make sure that the image that we find is actually the image that we’re looking for,

You can’t find a similar picture for me,

For example: I love you.jpg, I love him.jpg,

We’re looking for I love you.

In terms of features, SIFT has a 66% chance of finding I love him.

And SIFT thinks this is what we want, but it’s not,

We’re looking for I love you.

That’s why I added the histogram comparison, right

Quotes.

Thinking is the most important, other Baidu, Bing, StackOverflow, Android documents, autoJS documents, and finally in the group to ask — fang Shu tutorial

The statement

This tutorial is intended for learning purposes only and is not intended for any other use