Image processing large-scale popular science--image histogram

Image processing large-scale popular science--image histogram

Occasionally with the University of Zhejiang College of the lake freshman talk about contour recognition, she is doing a clothes can be classified into different styles of application, clothing evenly divided, and then with a mobile phone sweep, can tell you this is T-shirt or skirt, what? Do you dare say this thing is chicken? T-shirt and skirts are of course easy to separate, but those

Lotus skirt, dress, jumpsuit, half skirt, braces skirt, jumpsuit, hot pants, shorts, shorts, X-pants, leggings, boot pants, suspenders you really know?

The first to go aside to punish a song

Keep talking.

Learning elder sister probably do, first grayscale, then obtain high-frequency components, and then based on the high-frequency component binary, and then get the point set of the contour, and then through the histogram with the database of the preset model data to match the similarity, the final matching degree is the result of high.

Wait, histogram can make a match??

Words just after the yards, learn elder sister brush me a piece of English paper: "Look together?" (the link of the thesis dumps this: shape-context.pdf)

In the degree Niang, Ceres and Youdao brother's help to understand the meaning of the text, but also let me have a deeper understanding of the concept of the histogram

For just contact with MATLAB image processing, a talk about the histogram immediately think of using imhist () to draw the histogram of the image

or use the Histeq () balanced histogram, so that the picture contrast higher, and even remove the white noise in the picture (before also wrote an article to remove Haze White noise: matlab image processing: One minute to remove the haze in the picture)

Although the histogram in MATLAB has a lot of ready-made applications, but do not limit the definition of the histogram in these functions, it is difficult for others word\excel histogram is not it?

Lenovo to a variety of reports, copy of the histogram, I think we use the histogram, mainly want to have the function of classification and statistics,
For the histogram in image processing, the classification and statistics of Pixels is one of the most commonly used methods.

Then we get the histogram of gray value:

There are a lot of places to be used for this picture.

For example, we can use the slender peaks of the image to determine which grayscale the main information of the image is centered on.

Using the area between the two peaks of the image to determine which grayscale the background color is,

You can also use Peak Valley to determine the gray level of the noise.

It is also possible to use the grayscale expectation in the graph to give an ideal value of two values.

Even in hand-painted (recently a little fascinated by the painter hand-painted ... Determines the color range of the main color, background color, and highlight of the painting, greatly reducing the difficulty of color matching.

If this histogram is quantized, into a size = 256 of [Gray_value, Count_num] form of the matrix, it can be used as a unique fingerprint of this image, a 256-dimensional vector, no matter how many degrees of rotation of this picture, this vector will not change, yes, Finally around, any picture can correspond to such a vector, and the dimensions are 256 oh, as for the vector, how many dimensions of his, are a straight line, then the similarity between the two lines is not the angle of the straight line, (here do not put the formula, want to explore what the Baidu "n-dimensional vector angle") haha, This is not the perfect solution to the picture before and after the matching problem of rotation ~ ~

Work to be rigorous, I really think this is the histogram of the matching way, but carefully think, 256-dimensional vector angle .... This amount of calculation, this precision, but also to withstand the impact of environmental light, so harsh matching environment used in the identification of clothing style, almost no return

Back to the above mentioned in the paper, but also blame our understanding of the histogram is too superficial narrow, we have been in the histogram of gray values spinning, with the idea of statistics to touch the probability, but the histogram is not only statistics, but also can be classified, not only the histogram of gray values, but also what the histogram ah.

High energy ahead, focus attention

Specifically mentioned here is a histogram for classifying contour point sets :

A and B are the set of contour points of two graphs, C is a coordinate system, a circle is divided into 60 blocks according to the distance from the center of the circle, and the angle of North two, the angle is divided into 12 categories, the distance is divided into 5 classes, 12*5=60, and then each point of the point set is traversed to the center of C coordinate Divide the other points by 60 regions into a group of size = 60, preserving the number of contour points in each block, and now each point corresponds to a histogram of other points, like D, E, F three charts, 60 small squares, the more points, the darker the color.

The last step, compare A, b 22 points of similarity, to obtain the highest matching degree of the point and corresponding, the final result of the operation as shown in Figure g, no matter the image rotation, scaling, or the angle of the source of strength, will not have an effective effect on the results, to the maximum extent guaranteed matching accuracy.

This is a good example of the histogram in the classification of the special ability, I think in the subsequent application, we can also make the classification of the index, the abstract data through the histogram into a large class, and then a simple statistical operation, maximize the use of the histogram in the field of graph-phase processing classification statistics.

I hope that through this blog post can be around the things have more wonderful use of ideas and understanding of the way, always use other people's things also boring, anyway, this foreigner's thinking I was served

over~~

Image processing large-scale popular science-image histogram