阅读说明：本技术 一种基于边缘检测的电视台台标检测方法 (Television station logo detection method based on edge detection ) 是由 徐杰 谢恩鹏 刘永辉 于 2020-03-26 设计创作，主要内容包括：本发明提出一种基于边缘检测的电视台台标检测方法。本发明包括数据训练模块、图像获取模块、图像预处理模块、台标识别模块,其方法包括：获取各电视台台标图片；对数据集中台标图片进行灰度处理及边缘检测处理；构建卷积神经网络,获取台标检测模型；将台标检测模型下发至台标检测前端,将视频实时帧画面输入模型,返回该实时帧画面中是否包含台标及台标类别。本发明通过对数据集图片进行灰度处理及边缘检测处理,在保留台标图片特征的前提下减小图像原始数据量,将图片像素点转化为一维向量,有效减少了数据训练过程和识别过程中的计算量。本发明可以有效提高台标实时检测过程的效率。(The invention provides a television station logo detection method based on edge detection. The invention comprises a data training module, an image acquisition module, an image preprocessing module and a station caption identification module, wherein the method comprises the following steps: acquiring a logo picture of each television station; carrying out gray level processing and edge detection processing on the station caption picture in the data set; constructing a convolutional neural network to obtain a station logo detection model; and issuing the station caption detection model to a station caption detection front end, inputting the video real-time frame picture into the model, and returning whether the real-time frame picture contains the station caption and the station caption category. According to the invention, through carrying out gray level processing and edge detection processing on the data set picture, the original data volume of the picture is reduced on the premise of keeping the characteristics of the station caption picture, the picture pixel points are converted into one-dimensional vectors, and the calculation amount in the data training process and the identification process is effectively reduced. The invention can effectively improve the efficiency of the station caption real-time detection process.)

1. A television station logo detection method based on edge detection is characterized by comprising the following steps:

1) an image acquisition module of the detection center node intercepts a video frame picture played by a television station to acquire a station caption picture, and a station caption data set is constructed;

2) the image preprocessing module performs graying and edge detection processing on the station caption images in the station caption data set, acquires pixel points of the preprocessed images and converts the pixel points into one-dimensional vectors;

3) the data training module constructs a convolutional neural network, and performs multi-classification model training on the station caption data set in an off-line mode to obtain an optimal station caption detection model;

4) the detection central node issues the trained station caption detection model to a detection front end;

5) an image processing module at the front end of the detection is accessed into a detection video to intercept a station logo to be detected, and gray processing and edge detection processing are carried out on the station logo to be detected;

6) a station logo recognition module at the front detection end detects a target station logo by using a station logo detection model after offline training;

7) the detection front end transmits the recognition result and the detection picture back to the detection center node.

2. The method of claim 1, wherein the constructing the convolutional neural network comprises:

a) inputting the picture vectors into a convolutional neural network offline multi-classification model, and performing data normalization;

b) performing one-dimensional convolution pooling, multi-dimensional convolution pooling and full-connection layer processing on the station caption model, and judging whether the model is converged;

c) if the model is converged, the model is saved as the optimal model, if the model is not converged, the iterative processing is returned to the step a) to continue until the model is converged.

3. The method for detecting the TV station logo according to claim 1, wherein the method for detecting the edge of the logo picture is L aplace edge detection, L aplace operator has isotropy by using the second derivative information, i.e. it is not related to the direction of the coordinate axis, the gradient result is unchanged after the coordinate axis rotates, so that the image generates a steep zero crossing point at the edge after the second derivative, the edge is judged according to the zero crossing point, and L aplace operator defines that for a continuous function f (x, y) and its position (x, y) in the image:

the second order differential is negative on the bright side, positive on the dark side, the constant part is zero, the exact position of the side is determined, and whether the pixel is on the bright side or the dark side.

4. The method of claim 1, wherein there are one or more detection front ends.

5. The method of claim 1 wherein the station logo recognition module inputs the preprocessed frame pictures into the station logo detection model and returns whether the preprocessed frame pictures contain station logos and station logo types.

Technical Field

The invention relates to the technical field of digital broadcast television, in particular to a television station logo detection method and device based on edge detection in video images.

Background

With the development of the lcd of tv technology, the tv system in China has become a huge and complicated system, and the tv signals may be transmitted incorrectly, changed artificially and attacked maliciously in the course of front-end processing and network transmission, so that the supervision of tv signals is very important for ensuring the normal broadcast of digital tv programs, and the detection of station marks is an important way for supervising tv signals.

Conventionally, the station caption detection method is too powerful in work by monitoring television signals in real time in a manual mode and the like, the detection efficiency is low, and the problems of detection errors such as missing reports and the like exist. At present, station caption automatic detection usually utilizes deep learning to train a station caption detection model, and during detection, frame pictures are directly input into the model for identification, and because the picture quality is high, the real-time detection efficiency is low. The invention provides a television station logo detection method and device based on edge detection, which can effectively improve the efficiency of automatic station logo detection by preprocessing a detected picture, reducing the picture quality on the premise of keeping the detection picture characteristics.

Disclosure of Invention

In order to solve the problems of low station caption detection efficiency and high error rate, the invention provides the station caption detection method with high identification speed and high accuracy.

The invention discloses a television station logo detection method based on edge detection, which comprises the following steps:

1) an image acquisition module of the detection center node intercepts a video frame picture played by a television station to acquire a station caption picture, and a station caption data set is constructed;

2) the image preprocessing module performs graying and edge detection processing on the station caption images in the station caption data set, acquires pixel points of the preprocessed images and converts the pixel points into one-dimensional vectors;

3) the data training module constructs a convolutional neural network, and performs multi-classification model training on the station caption data set in an off-line mode to obtain an optimal station caption detection model;

4) the detection central node issues the trained station caption detection model to a detection front end;

5) an image preprocessing module at the detection front end is accessed into a detection video to intercept a station logo to be detected, and gray processing and edge detection processing are carried out on the station logo to be detected;

6) a station logo recognition module at the front detection end detects a target station logo by using a station logo detection model after offline training;

7) the detection front end transmits the recognition result and the detection picture back to the detection center node.

Preferably, the method for constructing the convolutional neural network comprises the following steps:

a) inputting the picture vectors into a convolutional neural network offline multi-classification model, and performing data normalization;

b) performing one-dimensional convolution pooling, multi-dimensional convolution pooling and full-connection layer processing on the station caption model, and judging whether the model is converged;

c) if the model is converged, the model is saved as the optimal model, if the model is not converged, the iterative processing is returned to the step a) to continue until the model is converged.

Preferably, the method for detecting the edge of the logo picture is L aplae edge detection, L aplae operator utilizes second derivative information to have isotropy, namely, the second derivative information is independent of the direction of a coordinate axis, the gradient result is unchanged after the coordinate axis rotates, so that an image generates a steep zero crossing point at the edge after being subjected to second derivative, the edge is judged according to the zero crossing point, and L aplae operator defines that for a continuous function f (x, y) and the position (x, y) of the continuous function f (x, y) in the image:

the second order differential is negative on the bright side, positive on the dark side, the constant part is zero, the exact position of the side is determined, and whether the pixel is on the bright side or the dark side.

Preferably, there are one or more detection front ends.

Preferably, the station caption identifying module inputs the preprocessed frame picture into the station caption detecting model and returns whether the frame picture contains the station caption and the station caption category.

Advantageous effects

According to the invention, through carrying out gray level processing and edge detection processing on the data set picture, the original data volume of the picture is reduced on the premise of keeping the characteristics of the station caption picture, the picture pixel point is converted into a one-dimensional vector, the calculated amount in the data training process and the identification process is effectively reduced, and the efficiency and the accuracy of the real-time station caption detection process are effectively improved.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.

FIG. 1 is a schematic diagram of the device architecture of the present invention;

fig. 2 is a schematic diagram of a convolutional neural network structure.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

A television station logo detection method based on edge detection is characterized in that a detection center node comprises a data training module for training a station logo detection model and obtaining an optimal model, an image obtaining module for obtaining a frame picture in a video stream, an image preprocessing module for performing graying and L aplace edge detection processing on an obtained original station logo, a station logo recognition module for comparing the obtained optimal station logo model with a station logo image detected in real time, an image obtaining module for obtaining a frame picture in the video stream, and an image preprocessing module for performing graying and L aplace edge detection processing on the obtained original station logo.

The following describes the implementation steps of the method in detail with reference to the accompanying drawings.

1) An image acquisition module of the detection center node intercepts a video frame picture played by a television station to acquire a station caption picture, and a station caption data set is constructed;

2) the image preprocessing module performs graying and edge detection processing on the station caption images in the station caption data set, acquires pixel points of the preprocessed images and converts the pixel points into one-dimensional vectors;

3) the data training module constructs a convolutional neural network, and the construction of the convolutional neural network comprises the following steps: inputting the picture vectors into a convolutional neural network offline multi-classification model, and performing data normalization; performing one-dimensional convolution pooling, multi-dimensional convolution pooling and full-connection layer processing on the station caption model, and judging whether the model is converged; if the model is converged, the model is saved as the optimal model, and if the model is not converged, the iterative processing is continued until the model is converged. Performing multi-classification model training on the station caption data set in an off-line manner to obtain an optimal station caption detection model;

4) the detection central node issues the trained station caption detection model to a detection front end;

5) an image preprocessing module at the detection front end is accessed into a detection video to intercept a station logo to be detected, and gray processing and edge detection processing are carried out on the station logo to be detected;

6) a station logo recognition module at the front detection end detects a target station logo by using a station logo detection model after offline training;

7) the detection front end transmits the recognition result and the detection picture back to the detection center node.

The method for detecting the edge of the station caption picture is L aplace edge detection, L aplace operator utilizes second derivative information to have isotropy, namely, the second derivative information is irrelevant to the direction of coordinate axes, the gradient result is unchanged after the coordinate axes rotate, a steep zero crossing point is generated at the edge after the image is subjected to second derivative, the edge is judged according to the zero crossing point, and L aplace operator defines that for a continuous function f (x, y) and the position (x, y) of the continuous function f (x, y) in the image:

the second order differential is negative on the bright side, positive on the dark side, the constant part is zero, the exact position of the side is determined, and whether the pixel is on the bright side or the dark side.

Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that changes may be made in the embodiments and/or equivalents thereof without departing from the spirit and scope of the invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.