Shulou(Shulou.com)06/02 Report--

This article is about how to use Python to build a search engine for image processing. The editor thinks it is very practical, so share it with you as a reference and follow the editor to have a look.

* companies and reverse image search

There are many tech companies that make good use of RIQ. For example, Pinterest implemented visual search for * times in 2014. Then in 2015 it released a white paper that revealed the structure of visual search. Reverse image search allows Pinterest to extract visual elements from fashionable things and then recommend similar products to consumers.

It is well known that Google images use reverse image search technology, which allows users to upload an image and then search for related images. The uploaded images are analyzed by algorithm to create a mathematical model, and then compared with thousands of images in the Google database until matching or similar images are found.

This is a picture of the OpenCV2.4.9 feature experience report.

Algorithm and Python Library

Before we get started, let's take a quick look at what Python elements are needed to build an image search.

Patent algorithm

SIFT (scale invariant feature transformation)

Cost-free patented technology that uses image identifiers to identify similar images. Even if the size, angle and depth of the picture are different, this technique can still match the image with similar content successfully.

SIFT uses a large number of features extracted from images as its search criteria.

It can match the pictures taken by the same object from different perspectives, and can search and match through the same characteristics.

SURF (accelerated robust feature) algorithm

SURF, like SIFT, is a free proprietary algorithm, which can be said to be an accelerated version of SIFT. Unlike SIFT, SURF uses box filters to approximate the Gauss Laplacian operator.

SURF uses the Hessian matrix to determine the location and size of the image.

In many applications, rotation invariance is not a necessary condition, so the algorithm is much faster without looking for direction.

SURF includes several features that speed up every step of the calculation. After the image has been rotated and blurred, SURF still works well and is three times faster than SIFT. However, SURF is not very good at changing light sources and viewing angles.

OpenCV provides SURF functionality, and SURF.compute and SURF.detect can be used to find descriptors and key points.

Open source algorithm

KAZE algorithm

KAZE is an open source 2D multi-scale and new feature detection and description algorithm, which operates in nonlinear scale space. The effective techniques of additive operator splitting (AOS) and variable thermal conductivity diffusion are used to construct nonlinear scale spaces.

The basis of multi-scale image processing is very simple-build a scale space of an image, and then use a function to process the image to enhance the scale or time of the image.

AKAZE (accelerated KAZE) algorithm

As you can guess from the name, this algorithm is faster, and it can quickly find the key to the match between two images. AKAZE uses a binary descriptor and a nonlinear scale space to balance accuracy and speed.

BRISK (binary robust scale invariant critical point) algorithm

BRISK is great at detecting and matching descriptors and key points.

This algorithm is very adaptable, scale space based on FAST, using bit-string descriptor detector, it can significantly speed up the search speed.

Key point detection and description in scale space are helpful to optimize performance.

FREAK (Fast Retinal key Point) algorithm

This is a new key point descriptor inspired by the human eye. A binary concatenated string is effectively calculated according to the image intensity. Compared with BRISK, SURF and SIFT,FREAK compute faster and save more memory.

ORB (native FAST and rotating BRIEF)

ORB is a binary descriptor that is noise resistant and rotation invariant. ORB is based on FAST key point detection and BRIEF descriptors, which together contribute to its low consumption and high performance.

In addition to fast and accurate orientation components, calculating native BRIEF and analyzing changes and relationships with native BRIEF are another major feature of ORB.

Python library

OpenCV

OpenCF is divided into academic version and commercial version. As an open source machine learning and machine vision library, OpenCV makes it easy for organizations to build their own versions based on OpenCV.

More than 2500 optimized algorithms, including the current machine learning and machine vision algorithms, can handle a variety of needs: face detection, object identification, camera movement tracking, finding similar images, eye tracking, scene recognition and so on.

* * companies like Google, Yahoo, IBM, Sony, Honda, Microsoft and Intel all use OpenCV libraries extensively.

OpenCV uses Python, Java, C, C++ and MATLAB as programming interfaces, and supports Windows, Linux, Mac OS and Android platforms.

Python Imaging Library (PIL)

PIL supports several language formats when doing image and graphics processing, which enables your Python to process images.

The standard process of image processing includes image enhancement, transparency and masking processing, image filtering, per-pixel operation and so on.

Build an image search engine

An image search engine can pick out similar images from existing images, which is Google's image search engine. For the new image search engine project, there are many ways to build, here are some:

An image search engine is established by using image extraction, image description extraction, metadata extraction and search result extraction.

Define your image descriptor, index the data set, define your similar reading matrix, and then search and rank.

Select the picture to search, select the folder of the existing picture, traverse the picture in the folder, create the picture feature index, match the same feature with the picture to search, and get the picture that can match.

Our approach is to match the grayscale version of the image at first, gradually start to use complex feature matching algorithms like SIFT and SURF, and finally use the open source algorithm BRISK. These algorithms only need a small modification to give effective results. Building an image search engine based on these algorithms can have many applications, such as analyzing the popularity of images, identifying objects in images, and so on.

Thank you for reading! This is the end of this article on "how to use Python to build a picture processing search engine". I hope the above content can be of some help to you, so that you can learn more knowledge. if you think the article is good, you can share it out for more people to see!

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.