Authenticity of Botero paintings. A study case.
Four easy steps to get a Botero artwork verified or appraised:
Is this "Woman on Green dress with Bird" painted by Fernando Botero?
The challenge
Problem description
We are commissioned to authenticate a painting attributed to Botero. For this work, we firstly created a dataset comprised of 100 photos of authentic signatures and artworks by Botero, verified by an art experts. Secondly, we painted a skilled forged signatures, to validate the result.
SIGNATURE ANALYSIS
Preparing the datasets
We converted the photos of signatures into squared B/W images, to use as input for our neural network, which works with sub-images of 256x256 pixels. 70% of the authentic signatures were separated for training, 10% for validation, and the remaining 20% were reserved for testing. We placed in a different folder the unknown signatures the customer provided us with.
STUDY OF THE ARTWORK
Study of the colors
To understand the color dynamics within the painting signed ”Botero - 92,” we’ve undertaken a color frequency analysis that aims to juxtapose the color usage in this piece against a collection of Fernando Botero’s verified works. The central premise of this investigation is to assess the stylistic congruence of painting A1 with Botero’s characteristic palette, especially focusing on the range and frequency of green hues used.
In our examination of Botero’s use of color, particularly green, we compiled a visual reference of 25 verified authentic oil paintings.
In the visual analysis of color frequencies, not only do the greens show a marked deviation in painting A1 from the ensemble of Botero’s verified works, but the reds and ochre tones follow a similar pattern. Subfigure 4a displays a vast array of these colors, reflecting the dynamism and variation in Botero’s approach when painting with a full spectrum. These reds and ochres are representative of the warmth and vibrancy often found in his compositions.
On the other hand, Subfigure 4b indicates that painting A1 utilizes a more constrained subset of these hues. The reds and ochres in A1 do not match the most probable shades that punctuate Botero’s authentic pieces. This inconsistency extends beyond the green palette, suggesting a broader divergence in color usage across the spectrum.
As with the greens, these findings regarding the reds and ochres should not be taken as definitive indicators of authenticity when viewed in isolation. Instead, they should be integrated into a more comprehensive evaluation process that considers the totality of Botero’s known practices, his artistic trajectory, and the specificities of each work.
The fact that A1’s color palette diverges from the statistical probabilities observed in the broader range of Botero’s works invites further investigative layers. It prompts questions about whether A1 represents an anomaly within his oeuvre due to experimental choices or whether the differences are attributable to another artist’s hand. Only through a meticulous and holistic examination that accounts for these color variations, alongside other critical factors such as provenance and stylistic elements, can a substantive determination regarding the authenticity of A1 be reached.
Training our neural network
Our neural networks are trained on the images of signatures
Tap on the blue box to read the complete report:
We employed our AI-based mathematical model, specially trained on digitized images of authentic signatures of Fernando Botero extracted from oil artworks dated in the same time period. These Admitted Signatures served as our foundation, supplemented with a dataset of Skilled Forgeries. To achieve peak classification accuracy, we optimized the images for robust feature map extraction. We set aside a subset of both authentic signatures and skilled forgeries as a control dataset, facilitating the validation of the model’s performance. The model underwent extensive training until its proficiency on the control dataset achieved a level considered adequate for dependable classification. Following this, we used the model to analyze signature Q1
The number of available authentic signatures is finite, so we perform data augmentation: This means randomly translating and rotating the shape of the signature in the squared images. This improves the accuracy of the classification. Additionally, it is also zoomed in and out a bit.
On the left hand side of this text, you can see on the second column, the signature after augmentation.
- We trained our model using only the authentic samples. For this task, we used two networks, an encoder and a decoder, which is able to reconstruct the signatures. After obtaining training convergence, we saved the trained model.
OUR VERDICT
Using our trained model to test the signatures
During testing we compute a probability of the signatures to be authentic.
- Technically, this is the similarity of the feature maps obtained with each test signature compared to the ones extracted during training with the authentic ones.
For testing, we used 50 fake and 50 authentic signatures, which were not seen by the model during training. As a result, of these, 99% of the signatures have been correctly classified.
CONCLUSION
We used A.I. and expert analysis to verify the artwork
To conclude, the signature and its adjacent area play an essential role in framing the artwork’s authenticity and origin. The unique character of the signature on A1, particularly the missing vertical stroke in the ”t” and the unusual presentation of a hyphen between the artist’s name and the year, coupled with the uncommon white color choice on a dark background, raises questions when compared to the Admitted Authentic Signatures provided for reference in Table 2. These anomalies suggest a deviation from Botero’s typical signature traits and warrant further scrutiny.
Need help to authenticate or value an item?
We offer a personalized service to our customers.
Four Easy Steps to Request Your Item's True Value:
If you find difficulties uploading the images, send them directly to [email protected] .
