CN117437570A - Underwater text recognition method, system, equipment and medium - Google Patents
Underwater text recognition method, system, equipment and medium Download PDFInfo
- Publication number
- CN117437570A CN117437570A CN202311308839.7A CN202311308839A CN117437570A CN 117437570 A CN117437570 A CN 117437570A CN 202311308839 A CN202311308839 A CN 202311308839A CN 117437570 A CN117437570 A CN 117437570A
- Authority
- CN
- China
- Prior art keywords
- underwater
- image
- dark channel
- channel
- identified
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000010586 diagram Methods 0.000 claims abstract description 42
- 238000012937 correction Methods 0.000 claims abstract description 36
- 238000004422 calculation algorithm Methods 0.000 claims description 78
- 238000000605 extraction Methods 0.000 claims description 22
- 238000002834 transmittance Methods 0.000 claims description 17
- 238000005286 illumination Methods 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 11
- 238000003384 imaging method Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 6
- 230000002708 enhancing effect Effects 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 4
- 239000013589 supplement Substances 0.000 claims description 2
- 230000011218 segmentation Effects 0.000 description 11
- 238000001514 detection method Methods 0.000 description 10
- 101100295091 Arabidopsis thaliana NUDT14 gene Proteins 0.000 description 9
- 230000015556 catabolic process Effects 0.000 description 6
- 238000006731 degradation reaction Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 6
- 238000011156 evaluation Methods 0.000 description 6
- 230000006872 improvement Effects 0.000 description 6
- 230000001965 increasing effect Effects 0.000 description 6
- 238000012360 testing method Methods 0.000 description 6
- 238000012545 processing Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000011176 pooling Methods 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 238000007781 pre-processing Methods 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 238000000889 atomisation Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000001502 supplementing effect Effects 0.000 description 2
- 241000251468 Actinopterygii Species 0.000 description 1
- 235000014653 Carica parviflora Nutrition 0.000 description 1
- 241000243321 Cnidaria Species 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000012015 optical character recognition Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000003014 reinforcing effect Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/09—Supervised learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/49—Segmenting video sequences, i.e. computational techniques such as parsing or cutting the sequence, low-level clustering or determining units such as shots or scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/30—Assessment of water resources
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Medical Informatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Character Input (AREA)
Abstract
The invention discloses an underwater text recognition method, an underwater text recognition system, underwater text recognition equipment and an underwater text recognition medium, which aim to solve the technical problem of low recognition efficiency of the existing underwater text recognition, and sequentially comprise the following steps: inputting a plurality of images to be identified; performing color correction on each image to be identified to obtain corrected images to be identified; obtaining a dark channel diagram of each pair of corrected images to be identified, and obtaining a dark channel diagram corresponding to each pair of images to be identified; judging whether the underwater environments in the image to be identified are the same or not according to the dark channel diagram corresponding to the image to be identified; if the underwater environments are the same, entering a step 5; if the underwater environments are different, entering a step 9; carrying out background light value estimation on a dark channel diagram in the same underwater environment; performing transmissivity estimation on the dark channel map; image enhancement is carried out on the dark channel diagram; performing character recognition on the dark channel graph to obtain a recognition result; the background light values are directly used for the dark channel map of the different underwater environment and then step 6 is entered.
Description
Technical Field
The present invention relates to the field of text recognition technology, and in particular, to a method, a system, an apparatus, and a medium for recognizing an underwater text.
Background
In recent years, with the development of ocean resources, professional imaging equipment is required to replace human self to collect information in most cases, such as submarine resource exploration, underwater salvage, submarine engineering and other engineering demands, so that the underwater image enhancement and character recognition are very important, and meanwhile, the underwater character recognition is also attractive with the increase of underwater robot operations. The rapid development of visual technologies such as deep learning, image processing and the like enables optical character recognition technology in various scenes to be researched and applied by a plurality of scholars.
Firstly, the traditional character recognition algorithm is only suitable for scenes which are easy to recognize, such as scenes with concise and clear fonts, high contrast between characters and the background, relatively simple background and the like. The preprocessing algorithm of the underwater original image also becomes a research hot spot in the research of character information recognition methods under complex natural backgrounds such as underwater character recognition and the like. Secondly, the difficulties of fuzzy contrast, distorted color, complex background and the like in the video or the image shot underwater lead to low character recognition efficiency, and the character recognition technology is also challenged.
The invention patent application with the application number of 201410699528.2 discloses a character recognition method of an underwater video image, which comprises the following steps: preprocessing the video image according to a morphological image processing principle, enhancing contrast and filtering noise; performing region segmentation on the video text according to the preprocessing result, and performing text region positioning by adopting a method based on the combination of edge detection and connected elements; according to the result of the video text region segmentation, a binary method combining a global threshold method and a local threshold method is adopted to segment the text, and the segmented characters are normalized, so that the segmented characters are consistent with the characters in a template library in size; and (3) designing a template library according to the characteristics of the video characters, matching the segmented characters with the characters in the template library, realizing character recognition and storing the characters in the text.
The character recognition method of the underwater video image adopts a mode of separating and processing region segmentation and character segmentation, the region segmentation is focused on the whole, the details are not focused, and the segmentation can be better realized without influencing the subsequent character segmentation; the character segmentation is further processed on the basis of the region segmentation, and the region segmentation can be realized faster than the conventional method aiming at character details, so that the character segmentation is realized better.
The character recognition method is the same as the prior art, and can realize character recognition of the underwater image, but the recognition efficiency of the underwater text is low due to the influence of the enhancement of the underwater degradation image, the complex underwater environment and the like in the underwater text recognition.
Disclosure of Invention
The invention aims at: the technical problem of low efficiency of the existing underwater text recognition is solved, and an underwater text recognition method, an underwater text recognition system, an underwater text recognition device and an underwater text recognition medium are provided.
The invention adopts the following technical scheme for realizing the purposes:
an underwater text recognition method comprising the steps of:
step 1, inputting a plurality of images to be identified;
step 2, carrying out color correction on each image to be identified to obtain a corrected image to be identified;
step 3, obtaining a dark channel diagram of each corrected image to be identified, and obtaining a dark channel diagram corresponding to each image to be identified;
step 4, judging whether the underwater environments in the images to be identified are the same according to the dark channel diagrams corresponding to the images to be identified; if the underwater environments are the same, entering a step 5; if the underwater environments are different, entering a step 9;
step 5, estimating a background light value of a dark channel diagram in the same underwater environment;
step 6, estimating the transmissivity of the dark channel map;
step 7, carrying out image enhancement on the dark channel map;
step 8, carrying out character recognition on the dark channel diagram to obtain a recognition result;
and 9, directly using background light values for dark channel diagrams of different underwater environments, and then entering step 6.
Further, step 2 includes red channel compensation and white balance when color correction is performed;
the red channel compensation is to correct the color cast by a red channel compensation algorithm;
white balance is the enhancement of white balance of underwater images according to Gray-Word hypothesis.
Further, the red channel compensation algorithm specifically comprises: the green channel information is used to supplement the red channel pixel RC, and the red channel pixel value is calculated as:
RC(i，j)＝R(i，j)+(G n -R n )×(1-R(i，j))×G(i，j)
wherein (i, j) represents each pixel location; r and G respectively represent the normalized results of the original red-green channel data; r is R n And G n Respectively representing the average value of red and green channel pixels;
the white balance enhancement is carried out on the underwater image according to Gray-Word hypothesis, specifically:
assuming that one RGB color image is subjected to color changes a plurality of times, the pixel average value of three channels tends to the gray level K:
K＝(RC+G+B)/3
wherein RC represents red channel pixels, and B and G represent the normalized results of the original blue-green channel data respectively;
the three-color channels of the same object under different illumination M and N are known to be represented asAndand a diagonal matrix containing gain coefficients of each channel is used for representing the relation of different illumination to the same object:
wherein K is r 、K g 、K b Respectively representing gain coefficients of three channels;
let λ ε { RC, G, B }, K λ ∈{K r ，K g ，K b Obtaining gain coefficient K of each channel of underwater image by K λ ：
K λ ＝K/λ，λ∈{RC，G，B}
Finally, obtaining the pixel value lambda of each channel of each enhanced image point after correcting the white balance color n ∈{R′，G′，B′}：
λ n ＝λ×K λ ，λ∈{RC，G，d}。
Further, in step 3, when the dark channel map is obtained, the dark channel map is obtained by using the image after color correction, and the calculation formula is as follows:
wherein y represents an image pixel point window, c represents RGB three channels, R ' G ' B ' represents pixel values of each channel of each enhanced image point after white balance color correction, and J c (y) represents one channel of R, G and B in the image; omega (x) represents the window where the image pixel is located at x.
Further, in step 5, when estimating the background light value, the specific estimation method is as follows:
estimating an underwater background light value A by selecting the maximum pixel point of the front 0.1% of the dark channel diagram obtained after color correction w 。
Further, in the step 6, when the transmissivity is estimated, the specific method is as follows:
by using the value A of the underwater background light w The transmittance is estimated, and the estimation formula is:
wherein y represents an image pixel point window, c represents RGB three channels, omega (x) represents a window where the image pixel point is located x, I c (y) represents a true blurred image obtained by the imaging system, A c Represents the underwater background light, J n (x) Representing a dark channel map; ω is the introduced coefficient, ω=0.95.
Further, step 8, performing character recognition on the dark channel map, and performing recognition by adopting a feature extraction network model;
the loss function of the feature extraction network model is:
L F ＝-α t (1-p t ) γ log(p t )
wherein y=1 means that the frame selected sample is a positive sample; p represents the probability that the sample is predicted to be classified for this; alpha t And gamma both represent balance factors, p t Representing the probability of sample prediction.
An underwater text recognition system comprising:
the image input module is used for inputting a plurality of images to be identified;
the color correction module is used for carrying out color correction on each image to be identified to obtain a corrected image to be identified;
the dark channel map acquisition module is used for acquiring the dark channel map of each pair of corrected images to be identified and obtaining a dark channel map corresponding to each pair of images to be identified;
the underwater environment judging module is used for judging whether the underwater environments in the images to be identified are the same according to the dark channel diagrams corresponding to the images to be identified; if the underwater environments are the same, entering a background light value estimation module; if the underwater environments are different, entering a background light value assignment module;
the background light value estimation module is used for estimating the background light value of the dark channel diagram in the same underwater environment;
the transmissivity estimation module is used for estimating the transmissivity of the dark channel graph;
the image enhancement module is used for enhancing the image of the dark channel map;
the character recognition module is used for carrying out character recognition on the dark channel graph to obtain a recognition result;
and the background light value assignment module is used for directly using the background light value for the dark channel diagrams of different underwater environments and then entering the transmissivity estimation module.
A computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of the method described above.
A computer readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of the method described above.
The beneficial effects of the invention are as follows:
in the invention, in the underwater character recognition algorithm, the input image is likely to be a continuous image of the same underwater environment according to the specific application scene of the algorithm, so that the estimation of the background light value is simplified in the algorithm, and the efficiency of the algorithm is improved. Meanwhile, when the background light value is estimated, the average of the background light values can be obtained by selecting the previous pictures to serve as the parameters of the water body, so that the robustness of the algorithm is improved. Multiplexing of the background light values enables the algorithm to have higher real-time performance in continuous character recognition.
Drawings
FIG. 1 is a schematic flow chart of the present invention;
FIG. 2 is a flow chart of a subsurface image enhancement algorithm of the present invention;
FIG. 3 is a schematic diagram of a feature extraction network model according to the present invention;
FIG. 4 is a view showing the contrast of the reinforcing effect of the underwater sunken ship, wherein (a) is an original drawing; (b) is a HE enhancement map; (c) is an SSR enhancement map; (d) is a DCP enhancement map; (e) is a UDCP enhancement map; (f) enhancing the graph by adopting the algorithm;
FIG. 5 is a comparative diagram of long text recognition, wherein (a) is a FOTS algorithm recognition diagram; (b) identifying a graph for the enhanced algorithm of the present application.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention.
Thus, all other embodiments, which can be made by one of ordinary skill in the art without undue burden from the invention, are intended to be within the scope of the invention.
Example 1
The embodiment provides an underwater text recognition method, which uses an improved underwater dark channel prior algorithm to enhance underwater images and uses an improved FOTS algorithm to recognize text. As shown in fig. 1, the method specifically comprises the following steps:
and step 1, inputting a plurality of images to be identified.
The acquired data set is a picture set marked with text labels.
The data sets used for underwater image enhancement effect evaluation are UIEBD data sets and for complex natures.
ICDAR 2015 dataset and ICDAR 2017MLT dataset trained and tested for text recognition under scene.
And 2, carrying out color correction on each image to be identified to obtain corrected images to be identified.
Performing color correction by compensating for a red channel and white balancing when performing color correction;
firstly, correcting color cast through a red channel compensation algorithm, and carrying out white balance enhancement on an underwater image according to Gray-Word hypothesis.
When the red channel compensation algorithm is performed, green channel information is used for supplementing red channel pixels RC, and the red channel pixel value calculation formula is as follows:
RC(i，j)＝R(i，j)+(G n -R n )×(1-R(i，j))×G(i，j)
wherein (i, j) represents each pixel location; r and G respectively represent the normalized results of the original red-green channel data; r is R n And G n Respectively representing the average value of red and green channel pixels;
the white balance enhancement is carried out on the underwater image according to Gray-Word hypothesis, specifically:
in an underwater environment, the primary reflected light is shifted from white light in the atmosphere to blue-green light. Therefore, after the red channel compensation, the white balance enhancement is also required for the underwater image according to the Gray-Word hypothesis. Specifically, assuming that one RGB color image is subjected to color change plural times, the pixel average value of three channels tends to the gradation K:
K＝(RC+G+B)/3
wherein RC represents red channel pixels, and B and G represent the normalized results of the original blue-green channel data respectively;
the three-color channels of the same object under different illumination M and N are known to be represented asAndand characterizing the same object by a diagonal matrix containing gain coefficients of each channelRelationship to different illumination:
wherein K is r 、K g 、K b Respectively representing gain coefficients of three channels;
let λ ε { RC, G, B }, K λ ∈{K r ，K g ，K b Obtaining gain coefficient K of each channel of underwater image by K λ ：
K λ ＝K/λ，λ∈{RC，G，B}
Finally, obtaining the pixel value lambda of each channel of each enhanced image point after correcting the white balance color n ∈{R′，G′，B′}：
λ n ＝λ×K λ ，λ∈{RC，G，B}。
The color shift caused by red channel information loss is compensated by the red channel compensated and white balanced image, and the white balance shifts the illumination condition mainly of the underwater blue light to the white light illumination so that the color of the image is more natural. The RGB three-channel pixels are distributed more naturally, and inaccurate dark channels caused by missing red channel information can be avoided in the subsequent dark channel diagram acquisition.
And step 3, obtaining a dark channel diagram of each pair of corrected images to be identified, and obtaining a dark channel diagram corresponding to each pair of images to be identified.
As the depth of field increases, the transmittance decays exponentially rapidly. The underwater propagation distances of light rays with different wavelengths are greatly different, the color shift can lead to the attenuation of pixel values of a red channel to be reduced, and an atomization area cannot be accurately identified. Therefore, the dark channel map is acquired by adopting the image after color correction in the step, and the calculation formula is as follows:
wherein y represents an image pixel point window, c represents RGB three channels, and R ' G ' B ' represents white balance color correctionPixel value of each channel of each enhanced image point after the enhancement, J c (y) represents one channel of R, G and B in the image; omega (x) represents the window where the image pixel is located at x.
Step 4, judging whether the underwater environments in the images to be identified are the same according to the dark channel diagrams corresponding to the images to be identified; if the underwater environments are the same, entering a step 5; if the underwater environment is different, the process proceeds to step 9.
And 5, estimating the background light value of the dark channel diagram in the same underwater environment.
In the step 5, when estimating the background light value, the specific estimation method is as follows: according to algorithm experience of dark channel prior, estimating an underwater background light value A by selecting the front 0.1% maximum pixel point of a dark channel diagram obtained after color correction w 。
And 6, estimating the transmissivity of the dark channel map.
In classical dark channel prior theory, it is assumed that the transmittance T (x) is a constant T w (x) And finally obtaining the transmissivity estimation. Will pass through J new Estimated underwater background light value A w Substituted formula
I c (x)≈(1-t(x))·A c ＝A c
The following formula is obtained:
further, the transmittance T can be obtained w (x) The formula is as follows:
in estimating the transmittance, since the newly obtained dark channel map is color mapped only by color correction and contains depth information of the image as the original image, the underwater background light value A obtained by the correction map can be used w The transmittance was estimated as follows. At the same time, the blurring phenomenon of omega restored real image distant view is introduced as the classical DCP algorithm, namely, estimationThe specific algorithm for calculating the transmissivity is as follows:
wherein y represents an image pixel point window, c represents RGB three channels, omega (x) represents a window where the image pixel point is located x, I c (y) represents a true blurred image obtained by the imaging system, A c Represents the underwater background light, J n (x) Representing a dark channel map; ω is the introduced coefficient, ω=0.95.
Through a large number of tests, setting ω=0.95 can increase the sense of realism of the image with a small amount of blur maintained.
After obtaining the underwater background light value A w And light transmittance T w (x) Then, through the formula of the underwater scattering model
Underwater restored image formula can be obtained:
T 0 to optimize the parameters. The transmittance obtained may be small, resulting in excessive pixel value of the restored image, so that T is introduced 0 Optimizing the formula according to the empirical chapter T 0 The value is 0.1.
And carrying out color correction on the original underwater image after red channel compensation and white balance, estimating a background light value and transmissivity, and finally restoring the defogged image through an underwater scattering model to obtain the underwater image enhanced by the chapter algorithm.
And 7, carrying out image enhancement on the dark channel map.
When the input pictures are images shot in different underwater environments, the input pictures can be regarded as a single picture task. Since underwater character images tend to be accompanied by various degrees of degradation of the underwater images, the recognition accuracy of the character recognition system is lowered, so that it is necessary to add an image enhancement algorithm for the underwater images (dark channel images).
Wherein the color correction of step 2, the dark channel of step 3, the background light value of step 5, the transmittance estimation of step 6, and the enhanced image of step 7, a underwater image enhancement algorithm is synthesized as shown in fig. 2.
And 8, carrying out character recognition on the dark channel diagram to obtain a recognition result.
And when character recognition is performed, a characteristic extraction network model is adopted for recognition. The feature extraction network model is a deep learning model for extracting useful features from raw data. There are drawbacks to long text recognition due to receptive field limitations in FOTS algorithms. Improvements are made to the shared convolutional network structure of feature extraction to incorporate a pyramid structure with hole convolution (Atrous Spatial Pyramid Pooling, ASPP). Conv5 in the shared convolution network structure is replaced by an ASPP network, a larger receptive field can be obtained in the ASPP network by changing expansion coefficients of layers of the pooling pyramid, and the characteristic extraction with a larger scale is realized, wherein the structure of the characteristic extraction network model is shown in figure 3.
The character recognition is improved aiming at the problems of insufficient receptive field of the FOTS algorithm and unbalanced sample training. Firstly, adding an ASPP structure into a FOTS algorithm feature extraction network to improve an algorithm receptive field; secondly, the standard cross entropy loss function of the algorithm is replaced and improved to be a Focal loss function, and the difficult positive sample training weight is increased.
The loss function of the improved feature extraction network model is as follows:
L F ＝-α t (1-p t ) γ log(p t )
wherein:
wherein y=1 means that the frame selected sample is a positive sample; p represents the probability that the sample is predicted to be classified for this; alpha t And gamma represents a balance factor, p t Representing the probability of sample prediction.
In the formula, the parameter gamma is [0,5 ]]. When p is t Approaching 1, the sample is a simple and easily divided sample, (1-p) t ) γ Approaching 0, at this time, the contribution of the simple easily separable sample to the loss value is reduced; when p is t Approaching 0, the sample is indistinguishable, (1-p) t ) γ And approaching to 1, the contribution ratio of the difficult-to-distinguish samples to the loss value is improved, so that the whole network is more concerned with the difficult-to-distinguish samples.
Finally at the loss function L F By a balance factor alpha t To suppress serious unbalance of the number of positive and negative samples, and to suppress unbalance of the number of simple and easily separable samples and indistinguishable samples by balancing factor gamma. In this example, alpha is selected experimentally t ＝0.25，γ＝2。
The underwater recognition algorithm mainly sets whether the water body environment is the same when a plurality of continuous images or video images are input. In the same underwater environment, the background light value in the underwater imaging model can be estimated only once, and the underwater character recognition algorithm is optimized.
When the input pictures are images shot in different underwater environments, the input pictures can be regarded as a single picture task. Since the underwater character image is often accompanied with various degrees of degradation of the underwater image, the recognition accuracy of the character recognition system is reduced, so that an image enhancement algorithm for the underwater image needs to be added.
When a plurality of continuous images converted from the same underwater environmental image or underwater video image are input, color correction and dark channel image acquisition are performed first. And then, only the background light value estimation is carried out on the first image, and the background light value of the first image can be directly used for the subsequent image due to the fact that the subsequent image has a close background light value under the same underwater environment. And then recovering the underwater image through the transmissivity graph and the background light value, and carrying out character recognition.
And 9, directly using background light values for dark channel diagrams of different underwater environments, and then entering step 6.
Test example 1:
in this embodiment, an underwater image of an underwater sunken ship is taken as an example for performing a test, the enhancement algorithm of this embodiment and other classical image enhancement algorithms are adopted for comparison, and the enhancement algorithms are selected as follows: histogram equalization algorithm, SSR algorithm based on Retinex model, dark channel priori defogging algorithm, underwater dark channel priori defogging algorithm and improved DCP algorithm.
The underwater image of the enhanced sunken ship is shown in fig. 4.
As shown in FIG. 4, the enhancement algorithm of the embodiment obtains better recovery of underwater artificial scenes such as underwater organisms including fish, coral and the like and underwater pipelines and the like in an underwater common complex scene. The improved algorithm effect diagram has the advantages that the red channel is compensated, the color is more uniform, the contrast is good, the overexposure and other problems are avoided, the problem of image degradation such as fog and blurring is solved, and good gain effects can be carried out on underwater images in different scenes.
TABLE 1 image enhancement algorithms PNSR, SSIM and UCIQE evaluation criteria score comparison Table
PSNR is the peak signal-to-noise ratio and is usually used as an evaluation index for the quality of signal reconstruction in engineering, so the same principle is also often used as objective evaluation of reconstruction such as picture compression or enhancement. SSIM comprehensively considers the quality of the enhanced picture from three aspects of brightness, contrast and structural similarity, and is close to subjective evaluation of people. The uci qe index evaluates underwater enhancement quality using a linear combination of contrast, saturation and chromaticity, describing color shift, blur and low contrast, respectively. As can be seen from table 1, the improved algorithm of this embodiment is only slightly lower than Retinex in PSNR index, but the PSNR algorithm only reflects the error and distortion of the image, which may not be consistent with the viewing gain effect of human eyes, and this index achieves a good effect. And is better than other algorithms in both other metrics.
Test example 2:
the images of the exhibition hall in this embodiment are tested in the ICDAR 2015 dataset for character recognition, and the test comparison result is shown in fig. 5.
As can be seen from fig. 5, the improved algorithm of the present embodiment improves detection of long samples by increasing receptive fields and detection of difficult positive samples. Most of text information in the picture is effectively detected by the algorithm before and after improvement, and the coordinates and the content are output. However, the FOTS algorithm has problems of insufficient recognition for characters with too large an aspect ratio difference, such as recognizing "recognition" as "cupunctus", or dividing the same word "root & Body" into "root" and "Body", and losing the small target "&" because of increasing recognition difficulty. Similar problems occur in the text TCM & message, the improved algorithm of the embodiment accurately performs text recognition, and the original algorithm is recognized in error due to the loss of the character @ and @, so that the recognition rate of the FOTS original algorithm is reduced. Through comparison of long text recognition, the improved algorithm of the embodiment is found to have good performance, the recognition accuracy of a long sample is effectively increased by increasing the receptive field in the feature extraction network, and long texts which cannot be detected in the original algorithm are accurately detected.
Table 2 results on ICDAR 2015 dataset versus table
Wherein S, W, G represent the three recognition tasks of strong, weak and generics in the data set respectively. P, R, F1 represent the precision, recall and F1 score in character detection. As can be seen from the experimental results table 2, the detection performance of the detection branch of the improved algorithm of the embodiment is obviously improved compared with that of the eats algorithm for FOTS reference, and the robustness of text detection is enhanced due to the introduction of the rotation operation on the characters. The improved algorithm of the embodiment exceeds the original algorithm in detection, and the improvement of the characteristic extraction network is proved to be beneficial to improving the detection comprehensive performance. Compared with FOTS, the improved algorithm obtains higher score in end-to-end recognition of text detection recognition, further illustrates that ASPP network is added in the feature extraction network aiming at the problem of long text recognition error detection, and truly improves the receptive field of the algorithm; for the problem of unbalance of positive and negative difficult samples in the data set, the improvement algorithm of the embodiment obtains the gain of identification accuracy for the improvement of the loss function.
Example 2
The embodiment provides an underwater text recognition system, which specifically includes:
and the image input module is used for inputting a plurality of images to be identified.
The acquired data set is a picture set marked with text labels.
The data sets used for underwater image enhancement effect evaluation are UIEBD data sets and for complex natures.
ICDAR 2015 dataset and ICDAR 2017MLT dataset trained and tested for text recognition under scene.
And the color correction module is used for carrying out color correction on each image to be identified to obtain a corrected image to be identified.
Performing color correction by compensating for a red channel and white balancing when performing color correction;
firstly, correcting color cast through a red channel compensation algorithm, and carrying out white balance enhancement on an underwater image according to Gray-Word hypothesis.
When the red channel compensation algorithm is performed, green channel information is used for supplementing red channel pixels RC, and the red channel pixel value calculation formula is as follows:
RC(i，j)＝R(i，j)+(G n -R n )×(1-R(i，j))×G(i，j)
wherein (i, j) represents each pixel location; r and G respectively represent the normalized results of the original red-green channel data; rn and Gn represent the average value of the red-green channel pixels, respectively;
the white balance enhancement is carried out on the underwater image according to Gray-Word hypothesis, specifically:
in an underwater environment, the primary reflected light is shifted from white light in the atmosphere to blue-green light. Therefore, after the red channel compensation, the white balance enhancement is also required for the underwater image according to the Gray-Word hypothesis. Specifically, assuming that one RGB color image is subjected to color change plural times, the pixel average value of three channels tends to the gradation K:
K＝(RC+G+B)/3
wherein RC represents red channel pixels, and B and G represent the normalized results of the original blue-green channel data respectively;
the three-color channels of the same object under different illumination M and N are known to be represented asAndand a diagonal matrix containing gain coefficients of each channel is used for representing the relation of different illumination to the same object:
wherein K is r 、K g 、K b Respectively representing gain coefficients of three channels;
let λ ε { RC, G, B }, K λ ∈{K r ，K g ，K b Obtaining gain coefficient K of each channel of underwater image by K λ ：
K λ ＝K/λ，λ∈{RC，G，B}
Finally, obtaining the pixel value lambda of each channel of each enhanced image point after correcting the white balance color n ∈{R′，G′，B′}：
λ n ＝λ×K λ ，λ∈{RC，G，B}。
The color shift caused by red channel information loss is compensated by the red channel compensated and white balanced image, and the white balance shifts the illumination condition mainly of the underwater blue light to the white light illumination so that the color of the image is more natural. The RGB three-channel pixels are distributed more naturally, and inaccurate dark channels caused by missing red channel information can be avoided in the subsequent dark channel diagram acquisition.
The dark channel map acquisition module is used for acquiring the dark channel map of each pair of corrected images to be identified and obtaining the dark channel map corresponding to each pair of images to be identified.
As the depth of field increases, the transmittance decays exponentially rapidly. The underwater propagation distances of light rays with different wavelengths are greatly different, the color shift can lead to the attenuation of pixel values of a red channel to be reduced, and an atomization area cannot be accurately identified. Therefore, the dark channel map is acquired by adopting the image after color correction in the step, and the calculation formula is as follows:
wherein y represents an image pixel point window, c represents RGB three channels, R ' G ' B ' represents pixel values of each channel of each enhanced image point after white balance color correction, and J c (y) represents one channel of R, G and B in the image; omega (x) represents the window where the image pixel is located at x.
The underwater environment judging module is used for judging whether the underwater environments in the images to be identified are the same according to the dark channel diagrams corresponding to the images to be identified; if the underwater environments are the same, entering a background light value estimation module; if the underwater environments are different, entering a background light value assignment module.
And the background light value estimation module is used for estimating the background light value of the dark channel diagram in the same underwater environment.
The specific estimation method of the background light value estimation module when estimating the background light value is as follows: according to algorithm experience of dark channel prior, estimating an underwater background light value A by selecting the front 0.1% maximum pixel point of a dark channel diagram obtained after color correction w 。
And the transmissivity estimation module is used for estimating the transmissivity of the dark channel graph.
False in classical dark channel prior theoryLet the transmittance T (x) be a constant T w (x) And finally obtaining the transmissivity estimation. Will pass through J new Estimated underwater background light value A w Substituted formula
I c (x)≈(1-t(x))·A c ＝A c
The following formula is obtained:
further, the transmittance T can be obtained w (x) The formula is as follows:
in estimating the transmittance, since the newly obtained dark channel map is color mapped only by color correction and contains depth information of the image as the original image, the underwater background light value A obtained by the correction map can be used w The transmittance was estimated as follows. Meanwhile, the blurring phenomenon of omega-restored real image distant view is introduced as the classical DCP algorithm, namely, the specific algorithm for estimating the transmissivity is as follows:
wherein y represents an image pixel point window, c represents RGB three channels, omega (x) represents a window where the image pixel point is located x, I c (y) represents a true blurred image obtained by the imaging system, A c Represents the underwater background light, J n (x) Representing a dark channel map; ω is the introduced coefficient, ω=0.95.
Through a large number of tests, setting ω=0.95 can increase the sense of realism of the image with a small amount of blur maintained.
After obtaining the underwater background light value A w And light transmittance T w (x) Then, through the formula of the underwater scattering model
Underwater restored image formula can be obtained:
T 0 to optimize the parameters. The transmittance obtained may be small, resulting in excessive pixel value of the restored image, so that T is introduced 0 Optimizing the formula according to the empirical chapter T 0 The value is 0.1.
And carrying out color correction on the original underwater image after red channel compensation and white balance, estimating a background light value and transmissivity, and finally restoring the defogged image through an underwater scattering model to obtain the underwater image enhanced by the chapter algorithm.
And the image enhancement module is used for carrying out image enhancement on the dark channel graph.
When the input pictures are images shot in different underwater environments, the input pictures can be regarded as a single picture task. Since underwater character images tend to be accompanied by various degrees of degradation of the underwater images, the recognition accuracy of the character recognition system is lowered, so that it is necessary to add an image enhancement algorithm for the underwater images (dark channel images).
And the character recognition module is used for carrying out character recognition on the dark channel graph to obtain a recognition result.
And when character recognition is performed, a characteristic extraction network model is adopted for recognition. The feature extraction network model is a deep learning model for extracting useful features from raw data. There are drawbacks to long text recognition due to receptive field limitations in FOTS algorithms. Improvements are made to the shared convolutional network structure of feature extraction to incorporate a pyramid structure with hole convolution (Atrous Spatial Pyramid Pooling, ASPP). Conv5 in the shared convolution network structure is replaced by an ASPP network, a larger receptive field can be obtained in the ASPP network by changing expansion coefficients of layers of the pooling pyramid, and the characteristic extraction with a larger scale is realized, wherein the structure of the characteristic extraction network model is shown in figure 3.
The character recognition is improved aiming at the problems of insufficient receptive field of the FOTS algorithm and unbalanced sample training. Firstly, adding an ASPP structure into a FOTS algorithm feature extraction network to improve an algorithm receptive field; secondly, the standard cross entropy loss function of the algorithm is replaced and improved to be a Focal loss function, and the difficult positive sample training weight is increased.
The loss function of the improved feature extraction network model is as follows:
L F ＝-α t (1-p t ) γ log(p t )
wherein:
wherein y=1 means that the frame selected sample is a positive sample; p represents the probability that the sample is predicted to be classified for this; alpha t And gamma represents a balance factor, p t Representing the probability of sample prediction.
In the formula, the parameter gamma is [0,5 ]]. When p is t Approaching 1, the sample is a simple and easily divided sample, (1-p) t ) γ Approaching 0, at this time, the contribution of the simple easily separable sample to the loss value is reduced; when p is t Approaching 0, the sample is indistinguishable, (1-p) t ) γ And approaching to 1, the contribution ratio of the difficult-to-distinguish samples to the loss value is improved, so that the whole network is more concerned with the difficult-to-distinguish samples.
Finally at the loss function L F By a balance factor alpha t To suppress serious unbalance of the number of positive and negative samples, and to suppress unbalance of the number of simple and easily separable samples and indistinguishable samples by balancing factor gamma. In this example, alpha is selected experimentally t ＝0.25，γ＝2。
The underwater recognition algorithm mainly sets whether the water body environment is the same when a plurality of continuous images or video images are input. In the same underwater environment, the background light value in the underwater imaging model can be estimated only once, and the underwater character recognition algorithm is optimized.
When the input pictures are images shot in different underwater environments, the input pictures can be regarded as a single picture task. Since the underwater character image is often accompanied with various degrees of degradation of the underwater image, the recognition accuracy of the character recognition system is reduced, so that an image enhancement algorithm for the underwater image needs to be added.
When a plurality of continuous images converted from the same underwater environmental image or underwater video image are input, color correction and dark channel image acquisition are performed first. And then, only the background light value estimation is carried out on the first image, and the background light value of the first image can be directly used for the subsequent image due to the fact that the subsequent image has a close background light value under the same underwater environment. And then recovering the underwater image through the transmissivity graph and the background light value, and carrying out character recognition.
And the background light value assignment module is used for directly using the background light value for the dark channel diagrams of different underwater environments and then entering the transmissivity estimation module.
Example 3
A computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of a method of underwater text recognition.
The computer equipment can be computing equipment such as a desktop computer, a notebook computer, a palm computer, a cloud server and the like. The computer equipment can perform man-machine interaction with a user through a keyboard, a mouse, a remote controller, a touch pad or voice control equipment and the like.
The memory includes at least one type of readable storage medium including flash memory, hard disk, multimedia card, card memory (e.g., SD or D interface display memory, etc.), random Access Memory (RAM), static Random Access Memory (SRAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), programmable Read Only Memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory may be an internal storage unit of the computer device, such as a hard disk or a memory of the computer device. In other embodiments, the memory may also be an external storage device of the computer device, such as a plug-in hard disk provided on the computer device, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), or the like. Of course, the memory may also include both internal storage units of the computer device and external storage devices. In this embodiment, the memory is often used to store an operating system and various application software installed on the computer device, such as program codes of the underwater text recognition method. In addition, the memory may be used to temporarily store various types of data that have been output or are to be output.
The processor may be a central processing unit (CentralProcessingUnit, CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor is typically used to control the overall operation of the computer device. In this embodiment, the processor is configured to execute the program code stored in the memory or process data, for example, the program code of the underwater text recognition method.
Example 4
A computer readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of a method of underwater text recognition.
Wherein the computer-readable storage medium stores an interface display program executable by at least one processor to cause the at least one processor to perform the steps of the underwater text recognition method as described above.
From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk), including several instructions for causing a terminal device (which may be a mobile phone, a computer, a server or a network device, etc.) to perform the underwater text recognition method according to the embodiments of the present application.
Claims (10)
1. An underwater text recognition method is characterized by comprising the following steps:
step 1, inputting a plurality of images to be identified;
step 2, carrying out color correction on each image to be identified to obtain a corrected image to be identified;
step 3, obtaining a dark channel diagram of each corrected image to be identified, and obtaining a dark channel diagram corresponding to each image to be identified;
step 4, judging whether the underwater environments in the images to be identified are the same according to the dark channel diagrams corresponding to the images to be identified; if the underwater environments are the same, entering a step 5; if the underwater environments are different, entering a step 9;
step 5, estimating a background light value of a dark channel diagram in the same underwater environment;
step 6, estimating the transmissivity of the dark channel map;
step 7, carrying out image enhancement on the dark channel map;
step 8, carrying out character recognition on the dark channel diagram to obtain a recognition result;
and 9, directly using background light values for dark channel diagrams of different underwater environments, and then entering step 6.
2. The method of claim 1, wherein step 2 includes red channel compensation and white balance when color correction is performed;
the red channel compensation is to correct the color cast by a red channel compensation algorithm;
white balance is the enhancement of white balance of underwater images according to Gray-Word hypothesis.
3. An underwater text recognition method as claimed in claim 2, wherein:
the red channel compensation algorithm specifically comprises the following steps: the green channel information is used to supplement the red channel pixel RC, and the red channel pixel value is calculated as:
RC(i，j)＝R(i，j)+(G n -R n )×(1-R(I，j))×G(i，j)
wherein (i, j) represents each pixel location; r and G respectively represent the normalized results of the original red-green channel data; r is R n And G n Respectively representing the average value of the red and green channel pixels.
The white balance enhancement is carried out on the underwater image according to Gray-Word hypothesis, specifically:
assuming that one RGB color image is subjected to color changes a plurality of times, the pixel average value of three channels tends to the gray level K:
K＝(RC+G+B)/3
wherein RC represents red channel pixels, and B and G represent the normalized results of the original blue-green channel data respectively;
the three-color channels of the same object under different illumination M and N are known to be represented asAndand a diagonal matrix containing gain coefficients of each channel is used for representing the relation of different illumination to the same object:
wherein K is r 、K g 、K b Respectively representing gain coefficients of three channels;
let lambda epsilon { RC, G, B },K λ ∈{K r ，K g ，K b obtaining gain coefficient K of each channel of underwater image by K λ ：
K λ ＝K/λ，λ∈{RC，G，B}
Finally, obtaining the pixel value lambda of each channel of each enhanced image point after correcting the white balance color n ∈{R′，G′，B′}：
λ n ＝λ×K λ ，λ∈{RC，G，B}。
4. An underwater text recognition method as claimed in claim 1, wherein: step 3, when obtaining the dark channel image, obtaining the dark channel image by adopting the image after color correction, wherein the calculation formula is as follows:
wherein y represents an image pixel point window, c represents RGB three channels, R ' G ' B ' represents pixel values of each channel of each enhanced image point after white balance color correction, and J c (y) represents one channel of R, G and B in the image; omega (x) represents the window where the image pixel is located at x.
5. An underwater text recognition method as claimed in claim 1, wherein: in the step 5, when estimating the background light value, the specific estimation method is as follows:
estimating an underwater background light value A by selecting the maximum pixel point of the front 0.1% of the dark channel diagram obtained after color correction w 。
6. An underwater text recognition method as claimed in claim 1, wherein: in the step 6, when the transmissivity is estimated, the specific method is as follows:
by using the value A of the underwater background light w The transmittance is estimated, and the estimation formula is:
wherein y represents an image pixel point window, c represents RGB three channels, omega (x) represents a window where the image pixel point is located x, I c (y) represents a true blurred image obtained by the imaging system, A c Represents the underwater background light, J n (x) Representing a dark channel map; ω is the introduced coefficient, ω=0.95.
7. An underwater text recognition method as claimed in claim 1, wherein: step 8, carrying out character recognition on the dark channel graph, and adopting a characteristic extraction network model for recognition;
the loss function of the feature extraction network model is:
L F ＝-α t (1-p t ) γ log(p t )
wherein y=1 means that the frame selected sample is a positive sample; p represents the probability that the sample is predicted to be classified for this; alpha t And gamma both represent balance factors, p t Representing the probability of sample prediction.
8. An underwater text recognition system, comprising:
the image input module is used for inputting a plurality of images to be identified;
the color correction module is used for carrying out color correction on each image to be identified to obtain a corrected image to be identified;
the dark channel map acquisition module is used for acquiring the dark channel map of each pair of corrected images to be identified and obtaining a dark channel map corresponding to each pair of images to be identified;
the underwater environment judging module is used for judging whether the underwater environments in the images to be identified are the same according to the dark channel diagrams corresponding to the images to be identified; if the underwater environments are the same, entering a background light value estimation module; if the underwater environments are different, entering a background light value assignment module;
the background light value estimation module is used for estimating the background light value of the dark channel diagram in the same underwater environment;
the transmissivity estimation module is used for estimating the transmissivity of the dark channel graph;
the image enhancement module is used for enhancing the image of the dark channel map;
the character recognition module is used for carrying out character recognition on the dark channel graph to obtain a recognition result;
and the background light value assignment module is used for directly using the background light value for the dark channel diagrams of different underwater environments and then entering the transmissivity estimation module.
9. A computer device, characterized by: comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of the method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized by: a computer program is stored which, when executed by a processor, causes the processor to perform the steps of the method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311308839.7A CN117437570A (en) | 2023-10-08 | 2023-10-08 | Underwater text recognition method, system, equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311308839.7A CN117437570A (en) | 2023-10-08 | 2023-10-08 | Underwater text recognition method, system, equipment and medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117437570A true CN117437570A (en) | 2024-01-23 |
Family
ID=89550728
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311308839.7A Pending CN117437570A (en) | 2023-10-08 | 2023-10-08 | Underwater text recognition method, system, equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117437570A (en) |
-
2023
- 2023-10-08 CN CN202311308839.7A patent/CN117437570A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11615559B2 (en) | Methods and systems for human imperceptible computerized color transfer | |
| Golts et al. | Unsupervised single image dehazing using dark channel prior loss | |
| KR102134405B1 (en) | System and Method for Improving Low Light Level Image Using Generative Adversarial Network | |
| US10984558B2 (en) | Learning-based sampling for image matting | |
| CN112288658A (en) | Underwater image enhancement method based on multi-residual joint learning | |
| CN110889813A (en) | Low-light image enhancement method based on infrared information | |
| US9311567B2 (en) | Manifold learning and matting | |
| CN109389569B (en) | Monitoring video real-time defogging method based on improved DehazeNet | |
| Feng et al. | URNet: A U-Net based residual network for image dehazing | |
| CN112836625A (en) | Face living body detection method and device and electronic equipment | |
| CN112749696A (en) | Text detection method and device | |
| CN111080754B (en) | Character animation production method and device for connecting characteristic points of head and limbs | |
| Wei et al. | An image fusion dehazing algorithm based on dark channel prior and retinex | |
| CN115033721A (en) | Image retrieval method based on big data | |
| CN117649694A (en) | Face detection method, system and device based on image enhancement | |
| CN114187515A (en) | Image segmentation method and image segmentation device | |
| CN116452469B (en) | Image defogging processing method and device based on deep learning | |
| CN117036202A (en) | Remote sensing image type imbalance-oriented hybrid enhancement method and system | |
| CN116798041A (en) | Image recognition method and device and electronic equipment | |
| CN113627342B (en) | Method, system, equipment and storage medium for video depth feature extraction optimization | |
| US20230186612A1 (en) | Image processing methods and systems for generating a training dataset for low-light image enhancement using machine learning models | |
| US20230186446A1 (en) | Image processing methods and systems for low-light image enhancement using machine learning models | |
| CN115761223A (en) | Remote sensing image instance segmentation method by using data synthesis | |
| Li et al. | Multi-scale fusion framework via retinex and transmittance optimization for underwater image enhancement | |
| CN117437570A (en) | Underwater text recognition method, system, equipment and medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |