November 4, 2026
⟢ Cloud Library: Find-Match V.2.1.0 – July 2026; stable
⟢ Comparative Protocol : COMP-V1 – 2024/2025 ; stable
For an optimal visual experience of the videos, kindly allow the buffering process to complete.
Mid-range prototype featuring a login interface. In the absence of valid credentials, the system restricts further analytical operations. This restriction is accompanied by a prompt for subscription. Should you be a corporate representative seeking to utilize this data blueprint design, arrangements can be made for the establishment of a bespoke user database.
Video image: extraction of a long vertical drop painting entitled Justice Pao aka. Bao Zheng (包青天); dimensions measuring 430 x 156 cm (height x width; 169.3 x 61.4 inches). This artwork was displayed at The National Art Museum of China in Beijing in 2025. The artist is Ma Shu Lin (马书林), created in 2018.
Once the images have been extracted, you may select any of the still image themes for analysis. For cloud-dependent searches, it is feasible to pursue image recognition matches from ART-Z within art paintings, moving images from ART-MOV, or your personal cloud domain.
This video offers a comprehensive understanding when contrasted with the COMP-V1 presented on the introductory page. The demonstration elucidates how users can upload specimens from diverse sources and in instances where the same specimen is viewed from multiple perspectives. For ceramics and porcelains, the theme enables cloud-dependent search functionality. In a manner akin to the VidG5 demonstration, users are afforded the opportunity to employ their customized cloud data library collection for the authentication process.
Users must be fully informed that, while COMP-V1 does not depend on cloud services, it requires the presence of two images to conclude the process: one target or test image and one reference image. When undertaking the authentication of an item, it is imperative to submit proof of the authentic product as the reference image for the authentication procedure. We strongly advise all users to commence the process utilizing a low-sensitivity setting as indicated on the interface. If a match-image is found in the cloud library, some information will be returned depending on the details in the descriptive file.
Images origin: Christie’s website
Art portraits full image search and segmental for validation.
For the purpose of still image recognition, please select the Art Portrait theme alongside your preferred cloud library. Within the library options, you have the opportunity to utilize Keras360 libraries for art images, images extracted from videos, and your own personalized cloud data. The interface will promptly initiate the image editor, provided that the uploaded image conforms to one of the supported formats.
Users are afforded the option to designate their session with a distinctive name, enabling them to resume their analysis at a subsequent time.
This serves as a secondary demonstration of VidG5. Given the duration of the video recording, the number of extracted frames is confined to the initial five seconds. The recording features a Be@rBrick that was launched in February 2022. All uploaded files are deemed temporary and will be removed if the user opts to reset or exit the session without saving. Additional information regarding our upload samples is available at the conclusion of this page.
A) User’s Data Image Library
Users are able to analyze image uploads in relation to their personal data image library. Please note that only the initial 50 images from the user’s library will be read and processed in this version.
B) Launch Image Editor
1) Image Editor: The full-sized image is uploaded and subjected to an automatic trimming process to eliminate excess background noise.
2) Image Editor: Manual adjustments are available for rotation, kurtosis (skewness), and distortion.
3) Image Editor: Image attributes provide information regarding the device type associated with the image’s origin.
C) Video-Splice
1) The video-splice feature to images (effective as of July 2026): Structured for analytical purposes.
2) In this iteration, the initial five seconds of uploaded video(s) will be converted into image frames.
3) A maximum of five image frames extracted from the video will concurrently operate during the authentication process.
D) Restore Previous Session
Call-me by My Name: In this version, users can save up to 5 sessions in their preferred names, and to restore the session to continue the analysis.
1) To initiate the process, it is essential to link the data library for authentication. This version incorporates the 2024/2025 protocols for the comparison of images in full-image search and segmentation. In this artistic image search, users may find it beneficial to edit their images utilizing the image editor, which supports PNG and JPG formats.
2) To launch the Image Editor, the size of the image in .jpg/.JPEG/.JPG format must not exceed 1MB.
3) Users may choose to preserve their session by entering their preferred name and saving it to the cloud server prior to exiting. The session can be restored at a later date.
4) Access to the features requires the use of registered login credentials.
5) Users should be cognizant of the size limitations regarding the processing of image data files.
1. In my personalised image library, should it only contain images?
No, it is not a requirement for the library to exclusively contain images. You may include a variety of file types. However, please note that only video and image file formats such as .MOV, .mp4, .JPG, .jpeg, .png, .tiff, and .heic will be accessible for retrieval.
2. Can I upload multiple images of the same specimen/object and attempt a search in the Cloud Library?
In the current iteration, this functionality is exclusively available during the VidG5 agent authentication process. Image frames extracted from a video clip are processed simultaneously to facilitate matching against the Cloud Library.
3. What do the parameters θ and Σ represent in the search outcome?
To accommodate variations in cropped image size and without a specific user preference regarding the cropping location on the image, this version implements a penalization mechanism during the image recognition process. Upon identifying a match, the search outcome retrieves the corresponding match-profile from the image library. Furthermore, it is important to note that there are technical parameters indicating the degree of similarities.
θ denotes the full-image resolution; the closer its value is to zero, the more favorable the outcome.
Σ signifies the similarities between the cropped image and the best-match image from the library. A general guideline for this parameter is that it should remain above zero. A strong image decompression may lead to a negative value.
4. I have come across a piece of artwork; however, I am uncertain whether it fully represents the original image or merely a fragment of a larger composition. Am I permitted to upload the image in the designated ‘full image’ field within the user interface?
Certainly, you may proceed accordingly. The protocol has been designed to accommodate both complete images and fragmentary representations. The system will initiate a validation process upon detecting a highly probable match during the image search. In cases where only a fragment of the original is uploaded, the AF agent will typically return a compilation of potential artists and corresponding artworks available in the database. In this demonstration, the audience will observe Rabbit-Watson gathering all possible outcomes in the dialog window.
If you are interested in 马书林 MaShuLin’s painting: Bao Zheng Justice Pao (2018) and the reference image available in the cloud library, you may refer to the images displayed below. The reference image does not need to be perfectly aligned; it may exhibit some imperfections at its corners, such as partial visibility of adjacent art portraits, reflective surfaces on the portrait, frames, and so forth. The calculator possesses a certain degree of tolerance to accommodate these aspects, provided they do not include excessive image kurtosis or distortions.
Artist. : Ma Shu Lin 马书林, creation 2018. Entitled : Justice Pao. Dimensions measuring 430 x 156 cm (height x width; 169.3 x 61.4 inches). Still image device : SONY (.JPG format). Video : IPhone (.MOV format)
The video was recorded on October 2nd, 2025, in Beijing, China, featuring MaShuLin’s artwork in .MOV format. In many real-world scenarios, tourists may inadvertently capture images or record video with their camera lens partially obstructed by objects such as a finger or a museum guide booklet.
This is an authentic BearBrick (size 1000%), standing at 70 cm tall, crafted from acrylic and owned by the founder of Keras360, Y Lin. BE@RBRICK Eugène Delacroix “Liberty Leading the People” is a release by Medicom Toy in Feb 2022, reflecting the events of the July Revolution of 1830 in France.
We are currently evaluating the scalability of this product. Should you be an investor interested in allocating resources toward significant development initiatives, we cordially invite you to connect with us at contact.general@keras360.io, or you may reach out to our founder, the principal architect behind the protocols, at nylin@naorococo.net.
🎵 Music
Ryo Yoshimata, Sound of Ocean
Tony Ann , Rain
Tony Ann, Ascension
© Keras360. All rights reserved.
Other relevant pages
