WO2022019747A1

WO2022019747A1 - System and method for identification of text region in video

Info

Publication number: WO2022019747A1
Application number: PCT/MY2020/050188
Authority: WO
Inventors: Hamam MOKAYED; Hock Woon Hon; Hasmarina HASAN; Shang Li YUEN
Original assignee: Mimos Bhd
Current assignee: Mimos Bhd
Priority date: 2020-07-24
Filing date: 2020-12-02
Publication date: 2022-01-27
Anticipated expiration: 2023-01-24
Also published as: MY209355A

Abstract

The present invention discloses a system and a method for identifying a text region in a video captured by a moving or still camera. The system comprises a video acquisition unit to obtain images recorded in an input video, an image processing unit comprising a seed point extraction unit and a text box determination unit to identify seed points of the images and locate potential text regions therefrom, and an inferencing unit comprising a classification module to characterize and verify the potential text regions. The seed point extraction unit comprises a statistical moment analysis module, a K-means clustering module, a N x N kernel for convolution module and a linearity evaluation module.

Description

SYSTEM AND METHOD FOR IDENTIFICATION OF TEXT REGION IN VIDEO

FIELD OF THE INVENTION

The present invention generally relates to the field of digital imaging. More particularly, the present invention relates to a system and a method for identifying a text region in a video captured using a camera.

BACKGROUND OF THE INVENTION

Text detection is important for any text recognition system where it deals with detecting presence of text in input images. It is one of the most challenging and commonly dealt in the field of artificial intelligence, augmented reality, and other innovations. The text may appear on any part of the input images and the regions that contain is referred to as text regions. The use of texts is extensive, and they can be found in many places, for example, car license plates, signboards, or billboards. These texts may be located or can be found at various positions which are difficult to access for scanning and information extraction.

For obtaining the input images, cameras on a moving platform such as unmanned aerial vehicle (UAV) and unmanned ground vehicle (UGV) may be employed. The cameras on the moving platform receive a good traction from video acquisition industry as this way of video acquisition leaves the cameras always in an active mode for reaching to many places. For instance, considers a car park management that requires recognition or identification of all vehicles’ license plates. Typically, a stationary, static or non-moving camera will be placed at entry and exit points of the car park in order to enable scanning of those license plates. This approach however is simply not feasible and practical for mass scanning as the stationary camera would only be able to scan one license plate at a time. The moving cameras provides a solution for mass scanning and information extraction. The moving cameras may be deployed to the car park for actively scanning all the parked vehicles.

The videos or images produced by the moving cameras however exhibits several unfavorable properties as compared to the static cameras. These properties include noises in terms of contrast, brightness, motion blur, perspective distortions and object quality in terms of level of details. Also, as the camera is moving in relative with the scene, it causes changes in height, distance, and angle which are directly affecting the view of the camera. The combination of these effects gives rise to low quality resultant video or image sequences. Besides that, due to the variability in imaging conditions, for example, lighting, specular reflections, commotion, obscure, and nearness of blocks over the content, and in the changeability of the content itself, for example, its scale, introduction, textual style, and style, the text detection remains a challenging problem.

In addition to the problems, as most of the time text detection is an uncooperative in nature and text can appear anywhere in the camera view, quite clearly, a new approach must be developed to detect the actual text regions of the texts before subsequence processes can be executed for recognition and other purposes.

By way of background, United States Patent 6,704,449 B1 discloses a method for extracting text from images whereby the intensity of each color plane of the image is identified. According to the ‘449 patent, the color plane with the highest intensity is converted into a black and white image, and character recognition is performed on the black and white image to determine the text of the image. The method of the ‘449 patent can be used for textual word searches for images with the capability to perform text searches for images. This method suggested by the ‘449 patent makes some global assumptions about background content that might work for displayed images such as those downloaded from web pages, but would have limited usefulness for videos and images that may have complex color backgrounds.

It would, therefore, be advantageous to provide a solution that would overcome the deficiencies and shortcomings of prior art by way of providing a system and a method for identifying the actual text regions in the videos captured using the moving cameras prior to text recognition or any other processes. Although there are systems and methods for the same in the prior art, for many practical purposes, there is still considerable room for improvement.

SUMMARY OF THE INVENTION

The following presents a simplified summary of the invention in order to provide a basic understanding of some aspects of the invention. This summary is not an extensive overview of the invention. Its sole purpose is to present some concepts of the invention in a simplified form as a prelude to the more detailed description that is presented later.

Accordingly, the present invention provides a system for identifying a text region in a video.

The system of the present invention comprises a video acquisition unit configured for receiving an input video from one or more cameras, wherein the video acquisition unit analyzes the input video to obtain images recorded thereof; The system further comprises an image processing unit configured for receiving the images transmitted from the video acquisition unit. The image processing unit comprises a seed point extraction unit configured for processing each pixel in the images to generate a phase congruency map having candidate pixels. The seed point extraction unit comprises a statistical moment analysis module configured to process the phase congruency map thereof to extract max moments of corner points associated with potential text regions; a K-means clustering module configured to cluster the max moments extracted thereof; a N x N kernel for convolution module configured to determine neighboring points in pair for each of the candidate pixels and distance values therebetween using the phase congruency map; and a linearity evaluation module configured to determine seed points of the potential text regions from the images by way of subjecting the distance values computed thereof to linearity checking. The image processing unit further comprises a text box determination unit comprising an edge detection module configured for generating edged images by way of processing the images using an edge detection model. The text box determination unit superimposes the seed points on the edged images to analyze edge data resulting therefrom. The text box determination unit determines a bounding box of the potential text regions based on the edge data thereof. The system further comprises an inferencing unit comprising a classification module configured for characterizing the potential text regions thereof against noises using a trained classifier model to determine a presence of text in the potential text regions, wherein the trained classifier model is selected from a group comprising a machine learning model, a Hidden Markov model, a support vector machine model, and a deep learning model.

Preferably, the one or more cameras include a moving camera and a still camera. Preferably, the seed point extraction unit determines, for each candidate pixel of the images, a local amplitude and phase and adopts the same for computing the phase congruency map.

Preferably, the statistical moment analysis module utilizes intensity information of the phase congruency map which represents phase and magnitude information per pixel to determine the corner points thereof.

Preferably, the edge detection model is selected from a group comprising a Laplacian filter, Gaussian filter, a Canny filter, an Otsu filter and a Sober filter.

Preferably, the text box determination unit assigns the edge data comprising nominated edges to edge groups based on geometric features which includes width, height, compactness, distance and slope, and removes any of the edge groups that has the nominated edges below a threshold.

Preferably, the potential text regions having the bounding box drawn thereof are an aggregation of the edge groups thereof.

Preferably, the system comprises a classifier model unit configured for preparing the trained classifier model and providing the same to the inferencing unit.

In accordance with another aspect of the present invention, there is provided a method of identifying a text region in a video.

The method of the present invention comprises the steps of receiving and analyzing an input video from one or more cameras to obtain images recorded thereof; processing each pixel in the images to generate a phase congruency map having candidate pixels; processing the phase congruency map thereof to extract max moments of corner points associated with potential text regions; clustering the max moments extracted thereof; determining neighboring points in pair for each of the candidate pixels and distance values therebetween using the phase congruency map; determining seed points of the potential text regions from the images by way of subjecting the distance values computed thereof to linearity checking; generating edged images by way of processing the images using an edge detection model; superimposing the seed points on the edged images to analyze edge data resulting therefrom; determining a bounding box of the potential text regions based on the edge data thereof; and characterizing the potential text regions thereof against noises using a trained classifier model to determine a presence of text in the potential text regions, wherein the trained classifier model is selected from a group comprising a machine learning model, a Hidden Markov model, a support vector machine model, and a deep learning model.

Further advantages of the present invention include that the system and the method enhance tourist decision-making, reduce users’ efforts and preserve their privacy, increase recommendation performance, and improve user satisfaction.

The foregoing and other objects, features, aspects and advantages of the present invention will become better understood from a careful reading of a detailed description provided herein below with appropriate reference to the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

A more complete appreciation of the invention and many of the attendant advantages thereof will be readily as the same becomes better understood by reference to the following detailed description when considered in connection with the accompanying drawings, wherein:

Figure 1 is a schematic diagram of a system for identifying a text region in a video according to one embodiment of the present invention;

Figure 2 is a schematic diagram of an image processing unit employed in the system of Figure 1 according to one embodiment of the present invention;

Figure 3 is a schematic diagram of an inferencing unit employed in the system of Figure 1 according to one embodiment of the present invention;

Figure 4 is a flow diagram of a method for identifying a text region in a video according to one embodiment of the present invention;

Figures 5A, 5B, 5C 5D, 5E and 5F, respectively, show a phase congruency map with candidate pixels, a first K-means processed image, a second K-means processed image, a N x N kernel for convolution, a first plot of distance values and a second plot of distance values for linearity checking which are related to the seed point extraction unit according to one embodiment of the present invention;

Figures 6A, 6B, 6C and 6D, respectively, show a processed image with extracted seed points, an edged image, a superimposed image and a filtered image which are related to the text box determination unit according to one embodiment of the present invention;

Figure 7 is a flow diagram of a classifier model employed in a classification module of an inferencing unit according to one embodiment of the present invention; and

Figures 8A and 8B, respectively, show a raw input image and a final processed image of a scene that is subjected to the system and method according to one embodiment of the present invention.

It is noted that the drawings may not be to scale. The drawings are intended to depict only typical aspects of the invention, and therefore should not be considered as limiting the scope of the invention.

DETAILED DESCRIPTION OF THE INVENTION

The present invention discloses a system and a method for identifying a text region in a video captured using one or more cameras prior to text recognition or any other processes.

The term “text region” as used herein refers to a region in an image that comprises text. Likewise, the term “potential text region” as used herein refers to a candidate text region in an input image that potentially contains text.

The one or more cameras essentially refer to an image capture device or multiple image capture devices and equivalents thereof known to those skilled in the art. The term “camera” may include moving, stationary, static, and still image cameras and video cameras. A camera may be analog or digital. In some embodiments, the camera may capture high-definition video. In some embodiments, the camera may be configured to stream image data (e.g. video data or files) to a computing device and/or remote computing device via a wired or wireless communication pathway. The computing device may be a camera or other memory storing the video data or files, a digital video recorder (DVR), a network video recorder (NVR) and the like.

According to one preferred embodiment of the present invention, the system comprises a video acquisition unit 100, an image processing unit 200, an inferencing unit 300 and a classifier model unit 400. An interconnection and interrelationship between the video acquisition unit 100, the image processing unit 200, the inferencing unit 300 and the classifier model unit 400 are schematically shown in Figure 1 .

The video acquisition unit 100 is preferably connected to the camera or to any computing device associated with the camera thereof. The video acquisition unit 100 is configured for receiving an input video streamed from the camera or the associated computing device. Upon receiving, the video acquisition unit 100 analyzes the input video thereof to obtain one or more images recorded thereof. It is preferred that the video acquisition unit 100 performs a conversion process onto the input video to generate images with an equivalent or appropriate resolution.

Once the images are completely extracted from the input video, the video acquisition unit 100 transmits those images to the image processing unit 200 for further processing. The image processing unit 200 preferably comprises a seed point extraction unit 201 and a text box determination unit 202 as shown in Figure 2.

The seed point extraction unit 201 is configured for processing each pixel in the images thereof to generate a phase congruency map having candidate pixels. It is also configured to determine, for each candidate pixel of the images, a local amplitude and phase and adopts the same for computing the phase congruency map.

The seed point extraction unit 201 preferably comprises a statistical moment analysis module 201a, a K-means clustering module 201b, a N x N kernel for convolution module 201c, and a linearity evaluation module 201 d. The statistical moment analysis module 201a is configured to process the phase congruency map thereof to extract max moments of corner points associated with potential text regions. In one embodiment, the statistical moment analysis module 201a utilizes intensity information of the phase congruency map which represents phase and magnitude information per pixel to determine the corner points thereof. In an alternative embodiment, the seed point extraction unit 201 may utilize any corner or edge detection models.

The K-means clustering module 201b is configured to cluster the max moments extracted from the phase congruency map thereof. In one embodiment, the K number can be any integer that is equal to or greater than 2.

The N x N kernel for convolution module 201c is configured to determine neighboring points in pair for each of the candidate pixels of the phase congruency map. It is further configured to compute distance values between the detected neighboring points using the phase congruency map thereof. In one embodiment of the N x N kernel for convolution module 201c, the N can be any odd integer, for examples 3, 5 and 7.

The linearity evaluation module 201 d is configured to determine, shortlist and finalize seed points of the potential text regions from the images. In one embodiment, the seed points can be obtained by way of subjecting the distance values computed thereof to linearity checking. The linearity checking involves a linearity evaluation to check all connected points are linear in any gradient value.

Once the seed points have been determined, the text box determination unit 202 will employ those seed points for a subsequent process. The text box determination unit preferably comprises an edge detection module 202a. It is configured for generating edged images by way of processing the images using an edge detection model. According to one embodiment, the edge detection model is selected from a group comprising a Laplacian filter, Gaussian filter, a Canny filter, an Otsu filter, and a Sober filter.

The text box determination unit 202 preferably superimposes the seed points on the edged images generated thereof. Edge data resulting therefrom which may include nominated edges will be subject to an analysis by the text box determination unit 202. Accordingly, the text box determination unit 202 assigns the edge data with the nominated edges to one or more edge groups based on geometric features. The geometric features may include width, height, compactness, distance, and slope or gradient. The text box determination unit 202 removes any of the edge groups that has the nominated edges below a predefined threshold. The edge groups which remains and were not removed by the text box determination unit 202 shall be considered in the following course.

At the next course, the text box determination unit 202 determines and draws a bounding box of the potential text regions based on the edge data thereof. It is preferred that the potential text regions having the bounding box drawn thereof are an aggregation of the remaining, non-removed edge groups thereof.

The inferencing unit 300 preferably comprises a classification module 301 as shown in Figure 3. It is configured for characterizing the potential text regions identified thereof against noises using a trained classifier model to determine a presence of text in the potential text regions. By performing this content classification, i.e. the determination of presence of text thereof, the inferencing unit 300 would be able to differentiate and hence, verify the said text regions in respect of text properties and non-text properties. In one embodiment, the inferencing unit 300 may be an online or offline inferencing unit.

The potential text regions employed at the inferencing unit 300 are preferably those that have been processed by the text box determination unit 202 and that have been drawn with the bounding box. In one embodiment, the trained classifier model is selected from a group comprising a machine learning model, a Hidden Markov model, a support vector machine model, and a deep learning model.

The classifier model unit 400 connected to the inferencing unit 300 is preferably configured for preparing the trained classifier model including training of classifier models using a set of training data to reflect salient attributes and behaviors of the phenomena being modeled. The trainings may include a parallel training of multiple classifier models or a sequential training of multiple classifier models. The training data includes historical training data associated with the identification of text regions in videos or images. The classifier model unit 400 is further configured to provide the trained classifier model to the inferencing unit 300. As with the method of the present invention, Figure 4 provides a summarized flow diagram depicting the steps involved, as explained in the preceding paragraphs in connection with the system thereof. For purposes of clarity in explanation and understanding, the method of the present invention will be described in the following section.

The method of the present invention preferably comprises the following steps:

5500 - receiving and analyzing an input video from one or more cameras to obtain images recorded thereof;

5501 - processing each pixel in the images to generate a phase congruency map having candidate pixels;

5502 - processing the phase congruency map thereof to extract max moments of corner points associated with potential text regions;

5503 - clustering the max moments extracted thereof;

5504 - determining neighboring points in pair for each of the candidate pixels and distance values therebetween using the phase congruency map;

5505 - determining seed points of the potential text regions from the images by way of subjecting the distance values computed thereof to a linearity checking;

5506 - generating edged images by way of processing the images using an edge detection model;

5507 - superimposing the seed points on the edged images to analyze edge data resulting therefrom;

5508 - determining a bounding box of the potential text regions based on the edge data thereof; and S509 - characterizing the potential text regions thereof against noises using a trained classifier model to determine a presence of text in the potential text regions,

The present invention will now be specifically described by the following examples, but it should be understood that the invention is not limited in any way to these examples.

Example 1

Determination of Seed Points

The determination of seed points is referred to the seed point extraction unit 201. For each pixel in an image, the seed point extraction unit 201 determines magnitudes of a local amplitude, A_no and phase, 0_no as defined in Equations 1 and 2, respectively.

By using the above equations, the seed point extraction unit 201 estimates a phase congruency for preparing a phase congruency map as defined in Equation 3 which is cosine minus magnitude of the sine of the phase deviation.

Figure 5A illustrates an exemplary phase congruency map of the present invention.

Further, the seed point extraction unit 201 utilizes information of phase congruency’s variation with orientation for use in the statistical moment analysis module 201a as defined in Equation 4 to calculate moment information including max moments.

The statistical moment analysis module 201a extracts the max moments of the detected points from the phase congruency map thereof (for corner points detection).

At the K-means clustering module 201 b, the pixels which represent a text cluster are classified using a K-means clustering model. In this regard, the K-means clustering model is set to value X. The value X can range from 2 to 5. Figures 5B and 5C illustrates an exemplary first K-means processed image and an exemplary second K-means processed image associated thereof.

Next, the K-means pixels are convoluted using 3 x 3 convolution by the N x N kernel for convolution module 201 c using information or values retrieved from the phase congruency map thereof, as follows:

Based on the convolution, the N x N kernel for convolution module 201c finds the closest neighboring points in pair (there will be eight points which form four start and end points to form a vector) and their distance values from each other. The N x N kernel for convolution module 201c calculates the distances values based on the closest neighboring points thereof. Figure 5D illustrates an exemplary convolution thereof. The linearity evaluation module 201 d checks the linearity of the distance values in order to find seed points. It formulates a line and links the four vectors into a plot. The linearity checking is as follows:

Figures 5E and 5F provides an exemplary first plot of distance values and an exemplary second plot of distance values for the linearity checking.

Eight neighboring points of each candidate pixel in the image will be used for detecting false candidates based on new clustering. The N x N kernel for convolution module 201c finds the minimum of its eight neighbors and then finds the element which is closest to it from other elements. This results in cluster-1 containing a pair of elements with close proximity. For the second iteration, it chooses the minimum from the list of the remaining elements (excluding those in cluster-1). This process results in four clusters containing pairs of elements for 8 neighbors of each candidate pixel. For each of the four clusters, the N x N kernel for convolution module 201c computes absolute differences, which outputs four distance values for the respective clusters. It is expected that if the candidate pixel happens to be a text pixel, the distances values must share nearly the same values and hence will satisfy the linearity property. Otherwise, the distance values satisfy the non-linearity property as illustrated in Figures 5E and 5F, where it can be seen that the line in Figure 5F behaves linearity for candidate text pixels and the line in Figure 5E behaves non-linearly property for false candidate pixels.

Example 2

Determination of Text Box

The text box determination unit 202 applies an edge detection model or algorithm onto a raw input frame, i.e. the images transmitted from the video acquisition unit 100 thereof to produce edged images.

The seed points in Example 1 are superimposed on or intersected with the edged images to obtain edge data which includes the nominated edges. The edges with overlapped seed points will be kept. The nominated edges (and neighboring edges) are grouped together to edge groups according to the geometric features such as width, height, compactness and distance.

The edge detection module 202a filters out the edge group(s) which has lower number of members according to the predefined threshold of number of edges. The isolated edge groups will be isolated. The edge detection module 202a aggregates the group elements to obtain the nominated text regions. Finally, a bounding box is provided and drawn based on the region of the edge groups.

Figures 6A, 6B, 6C and 6D, respectively, illustrate an exemplary processed image with extracted seed points obtained from the seed point extraction unit 201 , an exemplary edged image, an exemplary superimposed image and an exemplary filtered image.

Example 3

Content Classification

The inferencing unit 300 applies a trained classifier model to the potential text regions as identified by the text box determination unit 202. The trained classifier model subjects the potential text regions thereof through the flow diagram of the trained classifier model as shown in Figure 7. During the process, the classification module 301 of the inferencing unit 300 differentiates and verifies the text regions in respect of text properties and non-text properties to identify text object and non-text object.

With respect to the preparation of the trained classifier model, the classifier model unit 400 is employed. The classifier model unit 400 creates a trained classifier model with two classes that are capable of differentiating text properties from non-text properties, for instance text object and non-text object. The classifier model is trained with training data such as sample images of the said two classes.

Once trained, the classifier model unit 400 will apply the trained classifier model to the classification module 301 of the inferencing unit 300.

Figure 8A shows an exemplary raw input image of a scene. Figure 8B shows an exemplary processed image of the same scene with a plurality of bounding boxes drawn at the potential text regions detected thereof using the system and the method demonstrated by the present invention.

Although an overview of the inventive subject matter has been described with reference to specific example embodiments, various modifications and changes may be made to these embodiments without departing from the broader scope of embodiments of the present disclosure. Such embodiments of the inventive subject matter may be referred to herein, individually or collectively, by the term “invention” merely for convenience and without intending to voluntarily limit the scope of this application to any single disclosure or inventive concept if more than one is, in fact, disclosed.

The embodiments illustrated herein are described in sufficient detail to enable those skilled in the art to practice the teachings disclosed. Other embodiments may be used and derived therefrom, such that structural and logical substitutions and changes may be made without departing from the scope of this disclosure. The Detailed Description, therefore, is not to be taken in a limiting sense, and the scope of various embodiments is defined only by the appended claims, along with the full range of equivalents to which such claims are entitled.

As used herein, the term “or” may be construed in either an inclusive or exclusive sense. Moreover, plural instances may be provided for resources, operations, or structures described herein as a single instance. Additionally, boundaries between various resources, operations, modules, engines, and data stores are somewhat arbitrary, and particular operations are illustrated in a context of specific illustrative configurations. Other allocations of functionality are envisioned and may fall within a scope of various embodiments of the present disclosure. In general, structures and functionality presented as separate resources in the example configurations may be implemented as a combined structure or resource. Similarly, structures and functionality presented as a single resource may be implemented as separate resources. These and other variations, modifications, additions, and improvements fall within a scope of embodiments of the present disclosure as represented by the appended claims. The specification and drawings are, accordingly, to be regarded in an illustrative rather than a restrictive sense.

The foregoing description, for the purpose of explanation, has been described with reference to specific example embodiments. However, the illustrative discussions above are not intended to be exhaustive or to limit the possible example embodiments to the precise forms disclosed. Many modifications and variations are possible in view of the above teachings. The example embodiments were chosen and described in order to best explain the principles involved and their practical applications, to thereby enable others skilled in the art to best utilize the various example embodiments with various modifications as are suited to the particular use contemplated.

It will also be understood that, although the terms “first,” “second,” and so forth may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first contact could be termed a second contact, and, similarly, a second contact could be termed a first contact, without departing from the scope of the present example embodiments. The first contact and the second contact are both contacts, but they are not the same contact.

The terminology used in the description of the example embodiments herein is for the purpose of describing particular example embodiments only and is not intended to be limiting. As used in the description of the example embodiments and the appended examples, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the term “and/or” as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items. It will be further understood that the terms “comprises” and/or “comprising,” when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.

As used herein, the term “if” may be construed to mean “when” or “upon” or “in response to determining” or “in response to detecting,” depending on the context. Similarly, the phrase “if it is determined” or “if [a stated condition or event] is detected” may be construed to mean “upon determining” or “in response to determining” or “upon detecting [the stated condition or event]” or “in response to detecting [the stated condition or event],” depending on the context.

Claims

1. A system for identifying a text region in a video, characterized in that, the system comprising: a video acquisition unit (100) configured for receiving an input video from one or more cameras, wherein the video acquisition unit (100) analyzes the input video to obtain images recorded thereof; an image processing unit (200) configured for receiving the images transmitted from the video acquisition unit (100), comprising: a seed point extraction unit (201) configured for processing each pixel in the images to generate a phase congruency map having candidate pixels, comprising: a statistical moment analysis module (201a) configured to process the phase congruency map thereof to extract max moments of corner points associated with potential text regions; a K-means clustering module (201b) configured to cluster the max moments extracted thereof; a N x N kernel for convolution module (201c) configured to determine neighboring points in pair for each of the candidate pixels and distance values therebetween using the phase congruency map; and a linearity evaluation module (201 d) configured to determine seed points of the potential text regions from the images by way of subjecting the distance values computed thereof to linearity checking; and a text box determination unit (202) comprising an edge detection module (202a) configured for generating edged images by way of processing the images using an edge detection model, wherein the text box determination unit (202) superimposes the seed points on the edged images to analyze edge data resulting therefrom, wherein the text box determination unit (202) determines a bounding box of the potential text regions based on the edge data thereof; and an inferencing unit (300) comprising a classification module (301 ) configured for characterizing the potential text regions thereof against noises using a trained classifier model to determine a presence of text in the potential text regions.

2. The system according to Claim 1 , wherein the one or more cameras include a moving camera and a still camera.

3. The system according to Claim 1 , wherein the seed point extraction unit (201) determines, for each candidate pixel of the images, a local amplitude and phase and adopts the same for computing the phase congruency map.

4. The system according to Claim 1 , wherein the statistical moment analysis module (201a) utilizes intensity information of the phase congruency map which represents phase and magnitude information per pixel to determine the corner points thereof.

5. The system according to Claim 1 , wherein the trained classifier model is selected from a group comprising a machine learning model, a Hidden Markov model, a support vector machine model, and a deep learning model.

6. The system according to Claim 1 , wherein the edge detection model is selected from a group comprising a Laplacian filter, Gaussian filter, a Canny filter, an Otsu filter and a Sober filter.

7. The system according to Claim 1 , wherein the text box determination unit (202) assigns the edge data comprising nominated edges to edge groups based on geometric features which includes width, height, compactness, distance and slope, and removes any of the edge groups that has the nominated edges below a threshold.

8. The system according to Claim 6, wherein the potential text regions having the bounding box drawn thereof are an aggregation of the edge groups thereof.

9. The system according to Claim 1 further comprises a classifier model unit (400) configured for preparing the trained classifier model and providing the same to the inferencing unit (300).

10. A method of identifying a text region in a video, characterized in that, the method comprising the steps of: receiving and analyzing an input video from one or more cameras to obtain images recorded thereof (S500); processing each pixel in the images to generate a phase congruency map having candidate pixels (S501); processing the phase congruency map thereof to extract max moments of corner points associated with potential text regions (S502); clustering the max moments extracted thereof (S503); determining neighboring points in pair for each of the candidate pixels and distance values therebetween using the phase congruency map (S504); determining seed points of the potential text regions from the images by way of subjecting the distance values computed thereof to linearity checking (S505); generating edged images by way of processing the images using an edge detection model (S506); superimposing the seed points on the edged images to analyze edge data resulting therefrom (S507); determining a bounding box of the potential text regions based on the edge data thereof (S508); and characterizing the potential text regions thereof against noises using a trained classifier model to determine a presence of text in the potential text regions (S509), wherein the trained classifier model is selected from a group comprising a machine learning model, a Hidden Markov model, a support vector machine model, and a deep learning model.