System and method for image coding

Abstract

Relates to systems and methods for image coding. A method of encoding one or more images in a video includes: acquiring an original image from an image sensor of a video camera system; encoding the original image using the generated image model to obtain a first encoded image; decoding the first encoded image to obtain a first decoded image; identifying one or more regions of interest of the original image; for each of the regions of interest: performing a coding quality check by comparing a plurality of reference points in the region of interest of the original image with corresponding reference points in the region of interest of the first decoded image, thereby obtaining a difference level; if the difference level is greater than a threshold, encoding the region of interest using a non-generative image model, thereby obtaining a non-generative encoded image region; final encoded image data is provided for a first encoded image comprising a non-generated encoded image region having a region of interest with a difference level greater than a threshold and at least a remaining portion of the original image.

Description

Technical Field

[0001] The present disclosure relates to a method for encoding images in a video, and in particular, to a method for encoding images using artificial intelligence and performing quality checks thereof. The present disclosure further relates to an image processing system for implementing the disclosed method. Background Art

[0002] Image and video compression are widely used to reduce the cost of storage and compression. They usually work by applying algorithms to encode the image / video data during compression and decode the image / video data during playback or at the receiving data. Traditional compression usually involves many mathematical methods such as color space conversion, spatial compression, temporal compression, etc.

[0003] Recently, machine learning based approaches using, for example, convolutional neural networks and generative adversarial networks (GANs) have been developed. Artificial intelligence (AI) based image codecs, where trained neural networks perform encoding on the transmitter side and decoding on the receiver side, can provide high quality images / videos at significantly reduced bit rates. Using this technique, a simple representation of an image can be encoded by the network to output a description and / or multiple features that can be decoded by a network trained to reconstruct the encoded image.

[0004] In addition to significantly reduced bitrates, GANs are sometimes able to produce better images than traditional codecs, in the sense that they can produce something visually appealing even when there is very little information.

[0005] However, there are also known limitations and disadvantages associated with using machine learning for compression. "Hallusion" in the context of GAN encoding refers to incorrect features or details produced by the GAN during the encoding / decoding process. There may be many reasons for this behavior. Hallucination occurs when content is produced that was not present in the original data. In many applications, this is not a problem because the differences are insignificant or irrelevant to the specific use of the image data. However, if the compressed data is to be used as evidence and / or to be used in a trial or forensic investigation, there is sometimes an argument that data generated by artificial intelligence may have a lower evidentiary value. For example, the restored image may be contaminated by elements of the training data. Summary of the invention

[0006] The present disclosure relates to a method for encoding images in a video, and in particular, to a method for encoding images using artificial intelligence and preferably performing a quality check on the encoded images before transmission or storage. The final encoded image data may include both image data encoded using AI and image data encoded using a traditional codec.

[0007] A first aspect of the invention relates to a computer-implemented method for encoding one or more images in a video, the method comprising:

[0008] Acquire a raw image from an image sensor of a video camera system;

[0009] Encoding the original image using the generative image model to obtain a first encoded image;

[0010] Decoding the first encoded image to obtain a first decoded image;

[0011] identifying or obtaining one or more regions of interest of the original image;

[0012] For each of the regions of interest:

[0013] performing a coding quality check by comparing a plurality of reference points in the region of interest of the original image with corresponding reference points in the region of interest of the first decoded image, thereby obtaining a difference level;

[0014] If the difference level is greater than a threshold, the region of interest is encoded using a non-generative image model, thereby obtaining a non-generative coded image region;

[0015] Final encoded image data of the first encoded image is provided including a) a non-generated encoded image area of ​​the region of interest having a difference level greater than a threshold and b) at least a remaining portion of the original image.

[0016] It can be noted that the process of generating the final coded image data (i.e. the image data to be transmitted or stored) itself comprises both encoding and decoding as well as the steps of internally checking the quality of one or more regions of interest and then providing the final coded image data. The one or more regions of interest may be, for example, a face or a part of a face or a person or a specific part or detail on a building or a vehicle, but in principle may be any type of item or detail for a given application.

[0017] It can be said that the method is based on the distinction between parts of the original image that are critical to a particular application or use and parts of the original image that are less important, or at least for that application, it may be acceptable that the restored image may contain a certain degree of illusion. Therefore, the method includes the step of identifying or obtaining one or more regions of interest in the original image. As described above, the original image is both encoded to obtain a first encoded image and decoded to obtain a corresponding first decoded image. Then, for the identified one or more regions of interest, a quality check is performed in which a plurality of reference points in the original image are compared with corresponding reference points in the first decoded image to obtain a difference level for each region of interest. As will be appreciated by those skilled in the art, there is more than one way to determine the "difference level" of the region of interest. This is described in more detail below. As will also be understood by those skilled in the art, the expression "performing an encoding quality check" can also be regarded as a corresponding decoding check to some extent, because it is the first decoded image compared with the original image. A common practice in AI generative encoding is to train an encoder network and a decoder network in series. Once the process of evaluating the encoding has been completed, the method will decide whether the quality is good enough to use the image data that has been generated using the generative image model or whether encoding using a non-generative image model will have to be completed for the area. As a final step, final encoded image data of the first encoded image including a) a non-generated encoded image area of ​​the region of interest having a difference level greater than a threshold and b) at least a remaining portion of the original image is provided. The final encoded image data may include the first encoded image of only the remaining portion of the original image or more than the remaining portion of the original image (e.g., the entire original image).

[0018] One advantage of providing a flexibly coded image is that by using this arrangement it is possible to achieve a low bit rate for AI coding of the maximized image portion while minimizing the risk of introducing unacceptable illusions of key points or regions in the image.

[0019] The present disclosure further relates to an image processing system comprising:

[0020] The processing circuit is configured to:

[0021] Acquire a raw image from an image sensor of a video camera system;

[0022] Encoding the original image using the generative image model to obtain a first encoded image;

[0023] Decoding the first encoded image to obtain a first decoded image;

[0024] identifying or obtaining one or more regions of interest of the original image;

[0025] For each of the regions of interest:

[0026] performing a coding quality check by comparing a plurality of reference points in the region of interest of the original image with corresponding reference points in the region of interest of the first decoded image, thereby obtaining a difference level;

[0027] If the difference level is greater than a threshold, the region of interest is encoded using a non-generative image model, thereby obtaining a non-generative coded image region;

[0028] Final encoded image data of the first encoded image is provided including a) a non-generated encoded image area of ​​the region of interest having a difference level greater than a threshold and b) at least a remaining portion of the original image.

[0029] Those skilled in the art will recognize that the presently disclosed method of encoding one or more images in a video may be performed using any embodiment of the presently disclosed image processing system, and vice versa. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In the following, various embodiments are described with reference to the accompanying drawings. The drawings are examples of embodiments and are intended to illustrate some of the features of the presently disclosed system and method for encoding images in a video.

[0031] Figure 1 A flow chart illustrating a method according to an embodiment of the presently disclosed method of encoding one or more images in a video.

[0032] Figure 2 An example of a process of encoding an image according to an embodiment of the presently disclosed method is shown.

[0033] Figure 3 An example of comparison of multiple reference points in a region of interest of an original image with corresponding reference points in a region of interest of a decoded image is shown.

[0034] Figure 4 An embodiment of the presently disclosed image processing system is shown, wherein the presently disclosed method of encoding one or more images in a video is performed at a transmitter side, and wherein the final encoded image data is received at a receiver side. DETAILED DESCRIPTION

[0035] The present disclosure relates to a method of encoding one or more images in a video. Figure 1A flow chart of a method 100 according to an embodiment of the presently disclosed method for encoding one or more images in a video is shown. The method may include the step of acquiring one or more raw images from an image sensor of a video camera system (101). The acquired images may then be encoded using a generative image model to obtain a first encoded image (102). The method may further include the step of decoding the first encoded image to obtain a first decoded image (103). The decoding may be performed as part of an encoding process at the transmitter side. The method may further include the step of identifying or obtaining one or more regions of interest of the raw image (104). For each of the regions of interest, it is possible to perform a coding quality check by comparing a plurality of reference points in the region of interest of the raw image with corresponding reference points in the region of interest of the first decoded image, thereby obtaining a difference level for each region of interest. If the difference level is greater than a threshold for the region of interest, the region of interest can be encoded using a non-generative image model (105). The method may further include the step of providing final encoded image data of the first encoded image including a) a non-generative encoded image area of ​​the region of interest having a difference level greater than the threshold and b) at least a remaining portion of the original image (106).

[0036] It can be said that the "image model" includes a functional definition of an encoder (e.g., a functional specification, pseudocode, equations) and a functional definition of a decoder suitable for converting an image to and from a general image data format. The image data may be digital data. The generative image model may include one or more of a machine learning-based encoder, an AI-based encoder, an artificial neural network encoder, a generative adversarial network (GAN) encoder, a variational autoencoder (VAE), a convolutional neural network (CNN) encoder, and a recurrent neural network (RNN) encoder. If, for example, a GAN encoder is used, the first encoded image may be referred to as a GAN-encoded image, and the first decoded image may be referred to as a GAN-decoded image.

[0037] In contrast, using a non-generative image model can generally refer to the opposite, i.e., a more conventional way of encoding an image or video without using any of a machine learning-based encoder, an AI-based encoder, an artificial neural network encoder, a generative adversarial network (GAN) encoder, a variational autoencoder (VAE) encoder, a convolutional neural network (CNN) encoder, and a recurrent neural network (RNN) encoder to create information that is not in the original image or video. For example, encoding an image using a non-generative image model can be viewed as encoding the image without inserting information derived from images other than the encoded image. If the image is a frame of a video sequence, encoding the image using a non-generative image model can be viewed as encoding the image without inserting information derived from images other than the video sequence. Alternatively, the process of encoding an image using a non-generative image model can be viewed as encoding the image without processing the image by a function that depends on information derived from images other than the image (or, if the image is a frame in a video sequence, without processing the image by a function that depends on information derived from images other than the video sequence). Examples of non-generative image models include, but are not limited to, combinations of transform coding, predictive coding (e.g., inter-frame temporal predictive coding) and transform coding (so-called hybrid coding), ITU H.26x (particularly H.264, H.265, and H.266), AOMedia Video 1 (AV1), versions of JPEG such as JPEG 2000. At least H.26x and AV1 can be described as image models with hybrid coding.

[0038] Figure 2An example of a process 200 for encoding an image according to an embodiment of the currently disclosed method is shown. In the upper left corner, an original image 300 can be used as a starting point. The original image 300 includes a house 301a, a tree 301b, and a rabbit 301c, the latter representing an object of interest. By analogy, the rabbit may be a person, a face, or any other item of interest to the application. Starting from the original image 300, two parallel tasks are now performed. At this point, it should be understood that the system may include standard hardware blocks for performing various image processing tasks on the original image 300. For example, the system may include an image processing pipeline (IPP) for applying different filters such as noise filtering, segmentation, etc. At this point, the "original" image need not be considered as an unprocessed image, but rather as a starting point and input to the currently disclosed method. Needless to say, the camera has an image sensor that captures light, and the image sensor generates a signal that is processed to generate the original image. The tasks of the currently disclosed method do not have to be performed in any particular order, but may be performed in any suitable order. Preferably, the entire original image 300 is encoded using a generative image model 201. Then, the encoded image is decoded using the generated image model decoder 202. Thus, a first decoded image is obtained. Figure 2 As can be seen in FIG, the first decoded image is used in the encoding quality checker 204. The original image 300 may also be used in the process 203 for identifying one or more regions of interest 302 of the original image 300. In the step of performing the encoding quality check 204, the identified regions of interest of the original image 302a are compared with the regions of interest of the first decoded image 302b. The steps of the method may be arranged so that the identification of one or more regions of interest of the original image is performed before sending the first decoded image for quality check. Figure 2 In the example of , it can be seen that the "face" of the rabbit has been identified as a region of interest. In a quality check, a plurality of reference points in the region of interest of the original image are now compared with corresponding reference points in the region of interest of the first decoded image. Figure 3 This is further described and exemplified in . If at this point the difference between the original image 302a and the first decoded image 302b is too great (the difference level exceeds a threshold), the region of interest is encoded using a non-generated image model which may in practice be a conventional codec. In the step of providing / generating 205 final encoded image data 303, the final encoded image data is prepared such that it comprises a non-generated encoded image area of ​​the region of interest having a difference level greater than the threshold and a first encoded image of at least the remainder of the original image, i.e. potentially a mixture of image data generated using a generative image model and a non-generated image model.

[0039] The step of identifying one or more regions of interest of the original image may include detecting objects in the original image. The step of detecting objects may include applying a machine learning model such as a neural network that is trained to detect objects. Alternatively or in combination, the step of identifying one or more regions of interest of the original image may include defining one or more regions of interest based on the location of the detected object and / or what the object is doing and / or other contextual information (e.g., what is happening in the scene). The step of identifying one or more regions of interest of the original image is not limited to a specific way of identifying regions or objects, but actually includes multiple options. For example, this can involve additional neural networks that are trained to perform such tasks. As a non-limiting example, a semantic segmentation algorithm can detect in which (which) regions in the image there is grass, and decide that a car on the grass is an object of interest, while a car on the road is not considered to be of interest to a particular application.

[0040] Object detection is a computer vision technique that involves locating and identifying objects within an image frame. Those skilled in the art are generally familiar with such techniques and know how to implement them. Convolutional neural networks (CNNs) or other machine learning-based methods have gained popularity because they are often very accurate and fast, but there are many other object detection techniques that do not rely on CNNs or machine learning.

[0041] One specific non-limiting example of a more traditional type of object detection algorithm is the Viola-Jones detection framework. In this method, an image frame is scanned with a sliding window, where each region is classified as containing an object or not containing an object. This method uses Haar features and a cascade classifier for detecting objects.

[0042] When an object has been detected, a set of identification features can be created to describe the visual appearance of the detected object. The image data from a single image frame or video sequence can be used to create the identification features of the detected object. Various image and / or video analysis algorithms can be used to extract and create identification features from image data. The example of such image or video analysis algorithm is the algorithm for extracting features in the face in the "Face recognition using eigenfaces (face recognition using eigenfaces)" of Turk, Matthew A. and Alex P.Pentland in the CVPR'91 conference proceedings of the IEEE Computer Society "Computer Vision and Pattern Recognition 1991" conference in 1991, the IEEE 5th International "Automatic Face and Gesture Recognition 2002" conference proceedings of Lee Lily and W.EricL.Grimson in the "Gait analysis for recognition and classification (gait analysis for identification and classification)" for extracting gait features or various algorithms for extracting color in US 8,472,714 of Brogren et al.

[0043] The database may include a plurality of objects and a plurality of identification features. In the presently disclosed method, the step of identifying one or more regions of interest of the original image may include matching the identification features with the identification features in the database to classify the object as a certain type of object, such as a car, a person, or any other item. In one embodiment, the step of detecting the object in the image frame includes comparing the image frame with reference images in the database to match features corresponding to the object in the image frame and the reference image.

[0044] Classification of objects can be achieved through neural networks. Classification neural networks are often used in applications like character recognition, monitoring, surveillance, image analysis, natural language processing, etc. There are many neural network algorithms / techniques such as convolutional neural networks, recurrent neural networks, etc. that can be used to classify objects.

[0045] Figure 3 An example of comparing a plurality of reference points 304 in a region of interest 302a of an original image with corresponding reference points 304 in a region of interest 302b of a decoded image is shown. There are several options and ways of implementing a coding quality check of a region of interest. Coding quality checking is not limited to Figure 3 One option is to identify multiple key reference points for a given application. For example, Figure 3As can be seen in FIG. 1 , a plurality of feature points of a “face” have been identified. Accordingly, an embodiment of the currently disclosed method for encoding one or more images in a video may include the steps of identifying features of a given type of content in a region of interest and selecting reference points in the identified features. The identified features may be any type of item or feature, such as a nose and / or eyes or ears of a face and / or features of a person or features of a vehicle such as a license plate of a car and / or features of a building and / or features of an animal. The actual comparison may be, for example, a pixel-by-pixel comparison of a selection of pixels.

[0046] exist Figure 3 In the illustrative example of , multiple reference points have been identified. The identification of reference points can be accomplished by applying any machine learning model that has been trained to recognize certain features of an image. Those skilled in the art will generally be able to implement or use such a model.

[0047] Another option that can be used alone or in combination with the option of identifying specific reference points is to divide the region of interest into a grid, e.g. Figure 3 In the example shown in , the image is divided into a grid having cells 305. Reference points can then be distributed in the region of interest to form a grid. The grid can be a grid having cells that are uniformly sized and distributed or a grid having non-uniform cell sizes. For example, the method may include the step of using a function to create a grid having larger cells in an area that can be classified as background, and smaller cells in an area that can be classified as foreground or an area containing specific items or features. There are known methods that can, for example, distinguish between background and foreground in an image. Such methods will be readily available to those skilled in the art. Typically, the reference points can be selected so that the distance between reference points in the grid does not exceed a predetermined limit. In Figure 3 In the example in , the cells do not have to be uniformly sized. For example, there may be smaller cells in areas with more detail and / or more local variation and / or based on local content.

[0048] As described, the selection of reference points can be a combination of identifying features of a given type of content in the region of interest and dividing the region of interest into a grid of cells. In this combination, it is possible to have cells of smaller size in sub-regions close to the identified features and cells of larger size in the rest of the region of interest.

[0049] The currently disclosed method for encoding one or more images in a video may include a step of performing a coding quality check by comparing a plurality of reference points in a region of interest of an original image with corresponding reference points in a region of interest of a first decoded image, thereby obtaining a difference level. The "difference level" is introduced as a way to determine whether the encoded region of interest is close enough to the original image. It can be defined and measured in several ways. For example, pixels within a certain distance from a point or pixels in a cell of the above-mentioned grid can be compared between the region of interest of the original image and the region of interest of the first decoded image. This can be done pixel by pixel, or by grouping and averaging the pixels. As will be understood by those skilled in the art, different techniques and methods can be used. According to one embodiment of the currently disclosed method, a predefined measure of difference or dissimilarity can be used. An example is to calculate or extract the average pixel value of pixels within a certain distance from a point or pixels in a cell. If the difference between these values ​​is greater than a certain threshold, the quality of encoding the region of interest using a generative image model can be considered too low for the region of interest. Accordingly, then, a non-generative image model can be used to encode a specific region of interest. Another method would be pixel-to-pixel comparison. The sum of pixel differences or the sum of squared differences by accumulating pixel differences may be a further alternative. Another option may be to compare the histograms of the cells of the grid. In general, the step of obtaining the difference level is not limited to the specific examples provided in this disclosure.

[0050] Furthermore, in one embodiment of the presently disclosed method for encoding one or more images in a video, at least one of the regions of interest of the original image is composed of a plurality of image parameters such as color and brightness, wherein the step of performing the encoding quality check is performed only for a subset of the image parameters (e.g., only for brightness). For some applications, some image parameters may be irrelevant or less important. As an example, it may be important that the shape or other details of a face or person in the encoded image do not deviate too much from the original image, while other parameters such as color may not be as important.

[0051] The presently disclosed method of encoding one or more images in a video may include generating an aggregate encoding quality score for each region of interest. This information may be added to the final encoded image data for each region of interest. By adding the encoding quality scores for the regions of interest, a receiver will not only receive a certain minimum quality encoding, but also receive more detailed information about how similar it is to the original image. For example, if the similarity level is confirmed to be very high, the compressed data may ultimately be admitted as evidence and / or be accepted in a trial or forensic investigation.

[0052] The final coded image data may include a) a non-generated coded image area of ​​a region of interest having a difference level greater than a threshold and b) a first coded image of at least the remaining portion of the original image. Since the first coded image (i.e., the image encoded using the generative image model) will generally be compressed more, it is generally desirable to have as many first coded images as possible in the final coded image data. However, in areas where the difference between the first coded image and the original image is too significant, the final coded image may include a mixture of image data generated using the generative image model and the non-generated image model. Therefore, the step of decoding the final coded image data may include decoding the non-generated coded image area and decoding at least the remaining portion of the first coded image using a machine learning model such as a generative adversarial network without relying on information derived from images other than the non-generated coded image area. The final coded image data may include a first coded image of only the remaining portion of the original image or more than the remaining portion of the original image (e.g., the entire original image). If the final coded image data includes the first coded image of the entire image and a non-generated coded image area of ​​a region of interest having a difference level greater than a threshold, then selecting the data to be used in further processing may be a task of the receiving side. The receiving side will then generally take into account the information specifying the region of interest.

[0053] The present disclosure further relates to a computer program having instructions which, when executed by a computing device or a computing system, cause the computing device or the computing system to implement any embodiment of the method of encoding one or more images in a video disclosed herein. The computer program may be stored on any suitable type of storage medium, such as a non-transitory storage medium.

[0054] The present disclosure further relates to an image processing system comprising:

[0055] The processing circuit is configured to:

[0056] Acquire a raw image from an image sensor of a video camera system;

[0057] Encoding the original image using the generative image model to obtain a first encoded image;

[0058] Decoding the first encoded image to obtain a first decoded image;

[0059] identifying or obtaining one or more regions of interest of the original image;

[0060] For each of the regions of interest:

[0061] performing a coding quality check by comparing a plurality of reference points in the region of interest of the original image with corresponding reference points in the region of interest of the first decoded image, thereby obtaining a difference level;

[0062] If the difference level is greater than a threshold, the region of interest is encoded using a non-generative image model, thereby obtaining a non-generative coded image region;

[0063] Final encoded image data of the first encoded image is provided including a) a non-generated encoded image area of ​​the region of interest having a difference level greater than a threshold and b) at least a remaining portion of the original image.

[0064] The processing circuit or further processing circuits may be configured to decode the final encoded image data at the receiver side.It can be noted, therefore, that encoding may comprise both an encoding step and a decoding step in order to verify and / or improve the final encoded image data before transmitting or storing it. Figure 4 An example of such an embodiment is shown. An image processing system 500 comprises a processing circuit 501 configured to acquire an image from a camera system 502. The image processing system 500 in this arrangement may be considered as the transmitter side of the system, which may further comprise a receiver side system 400 having a processing circuit 401 configured to receive the transmitted compressed final encoded image data and decode it. Both the step of encoding the original image using the generative image model and the step of decoding the first encoded image as part of the encoding quality check may be performed on the transmitter side before transmitting the final encoded image data to the receiver side.

[0065] The image processing system may further include peripheral components such as one or more memories that can be used to store instructions that can be executed by any of the processors. The one or more memories may include random access memory (RAM) and / or read-only memory (ROM) or any suitable type of memory. The system may further include any of the following components: an internal network interface and an external network interface, an input and / or output port, a module for wirelessly sending and receiving data, a communication interface that allows software and / or data to be transmitted between the system and an external device. The software and / or data transmitted via the communication interface may be an electrical signal, an optical signal, or an RF signal in any suitable form. As will be appreciated by those skilled in the art, the processing circuit may be a single processor in a multi-core / multi-processor system.

Claims

1. A computer-implemented method for encoding one or more images in a video, the method comprising: Acquire a raw image from an image sensor of a video camera system; Encoding the original image using a generative image model to obtain a first encoded image; Decoding the first encoded image to obtain a first decoded image; Identifying or obtaining one or more regions of interest of the original image; For each of the regions of interest: identifying features of a given type of content in the region of interest and selecting a plurality of reference points within the identified features, and / or dividing the region of interest into a grid of cells, wherein a plurality of reference points are distributed in the region of interest to form the grid of cells; performing a coding quality check by comparing the reference points in the region of interest of the original image with corresponding reference points in the region of interest of the first decoded image, thereby obtaining a difference level; If the difference level is greater than a threshold, encoding the region of interest using a non-generative image model, thereby obtaining a non-generative coded image region; Final encoded image data of the first encoded image is provided comprising a) the non-generated encoded image area of ​​the region of interest having a difference level greater than the threshold value and b) at least a remaining portion of the original image.

2. The method according to claim 1, wherein: The generative image model comprises a generative adversarial network (GAN) and / or a variational autoencoder (VAE) encoder, and wherein the first encoded image is a GAN-encoded image and / or a VAE-encoded image, and the first decoded image is a GAN-decoded image and / or a VAE-decoded image.

3. The method according to claim 1, wherein: The reference points are selected such that distances between the reference points in the grid do not exceed a predetermined limit.

4. The method according to claim 1, wherein: The identified features include nose and / or eyes or ears of the face and / or features of the person, and / or wherein, The identified features include features of a vehicle such as a license plate of a car and / or features of a building and / or features of an animal.

5. The method according to claim 1, wherein: The step of identifying one or more regions of interest of the original image comprises detecting objects in the original image, preferably wherein the step of detecting objects comprises applying a machine learning model such as a neural network trained to detect the objects.

6. The method according to claim 1, wherein: The step of identifying one or more regions of interest of the original image includes defining the one or more regions of interest based on locations of detected objects.

7. The method according to claim 1, wherein: The at least one region of interest of the original image consists of a plurality of image parameters such as color and brightness, wherein the step of performing the encoding quality check is performed only for a subset of the image parameters, for example, only for the brightness.

8. The method according to claim 1, wherein: The step of performing the coding quality check comprises generating a total coding quality score for each region of interest, and preferably further comprises the step of adding the total coding score for each region of interest to the final encoded image data.

9. The method according to claim 1, wherein: Before transmitting the final encoded image data to the receiver side, the step of encoding the original image using the generative image model and the step of decoding the first encoded image are both performed on the transmitter side, preferably, further including the step of receiving the final encoded image data on the receiver side and decoding it.

10. The method according to claim 9, wherein: The step of decoding the final encoded image data includes decoding the non-generated encoded image area without relying on information derived from images other than the non-generated encoded image area and decoding the first encoded image of at least the remaining portion using a machine learning model such as a generative adversarial network.

11. A computer program having instructions, which, when executed by a computing device or a computing system, cause the computing device or the computing system to implement the method of encoding one or more images in a video according to claim 1.

12. An image processing system, comprising: The processing circuit is configured to: Acquire a raw image from an image sensor of a video camera system; Encoding the original image using a generative image model to obtain a first encoded image; Decoding the first encoded image to obtain a first decoded image; Identifying or obtaining one or more regions of interest of the original image; For each of the regions of interest: identifying features of a given type of content in the region of interest and selecting a plurality of reference points within the identified features, and / or dividing the region of interest into a grid of cells, wherein a plurality of reference points are distributed in the region of interest to form the grid of cells; performing a coding quality check by comparing the reference points in the region of interest of the original image with corresponding reference points in the region of interest of the first decoded image, thereby obtaining a difference level; If the difference level is greater than a threshold, encoding the region of interest using a non-generative image model, thereby obtaining a non-generative coded image region; Final encoded image data of the first encoded image is provided comprising a) the non-generated encoded image area of ​​the region of interest having a difference level greater than the threshold value and b) at least a remaining portion of the original image.

13. The image processing system according to claim 12, wherein: The processing circuit or further processing circuits are configured to decode the final encoded image data at a receiver side.

Images