AI heatmaps catch clever image theft and unauthorized sellers

It is obvious that they have used our brand's official images without authorization, but when we actually file a report with the platform, the response we get back is always the same. “Insufficient evidence.”

From the practitioner's perspective, it is absurd. This is because when you look at it with your own eyes, it is so obvious. For minor variations, such as slightly changing the background color or adding a watermark on top of the authentic image, the platform rejects the report on the grounds that “it is difficult to consider it identical to the original.” For the person in charge who always looks at the brand image, even small differences are bound to stand out, but platforms that are not familiar with our brand cannot easily distinguish small differences with the naked eye.

As such, practitioners have the conviction that 'it is indeed a theft.' However, because there are no objective criteria or figures to support this, the platform does not accept the report. Ultimately, the result is always the same. Brand images are being used without authorization, but sanctions to stop this fall through the cracks.

If it were a simple copy, it would be easy to filter out, but the problem lies in the fact that these 'clever modifications' neutralize the grounds for reporting. Therefore, practitioners hit the exact same wall every time. Frustrating questions like, “This is clearly unauthorized use, so why can't we prove it?” keep piling up.

A Visual Basis to Increase Report Success Rates: Heatmaps

Repeated “insufficient evidence” rulings cause confusion even within the brand. The person in charge is absolutely sure that it is a theft, but since there are no objective figures or visual materials to support this, when the question “Is there any basis to definitively conclude that it is a theft?” comes up during the internal reporting process, it is difficult for the person in charge to answer clearly. As a result, they have no choice but to keep letting the illegally distributed images slide, and consequently, unauthorized sellers continue to sell, and the damage to the brand grows. A bigger problem is that as these cases accumulate, an atmosphere of “nothing changes even if we file a report” is created within the brand, making it difficult for practitioners to respond actively.

Here, a heatmap becomes an important breakthrough. AI does not just look at images with human-like senses, but visually displays specifically which elements are the same and where the differences lie. For example, if the curve of a cosmetic bottle, the line spacing of label text, and the texture or tone of the package surface are identical, that area is displayed in blue on the heatmap. Conversely, areas with low similarity, such as deleted or synthesized areas, or areas with changed background colors, are highlighted in colors close to red, passing through yellow and orange. In other words, you can see at a glance that if there is a lot of blue, it is mostly identical to the original, and where red gathers, it is an area where differences have occurred. We will look at heatmap images in more detail below, along with specific practical application cases.

Thanks to this differentiation, practitioners no longer have to rely solely on “gut feeling.” They can explain the basis in a way such as, “This is an AI analysis result showing that the product bottle shape and label text position are identical,” which increases persuasiveness in internal decision-making as well. Furthermore, at the platform reporting stage, instead of simply claiming that “it is similar,” they can clearly explain “why this image can be seen as unauthorized use” by presenting visualized data.

Ultimately, a heatmap is a tool that translates the conviction felt by brand practitioners into an objective language. With this, a new system of evidence is established to overcome the walls that were repeatedly encountered during the reporting process.

Then, in what actual situations can a heatmap play this role? Let's look at specific practical cases to see how this tool is applied and what difference it makes for brands.

What if we apply heatmaps to our company's situation?

In our last content, we summarized the 7 clever modification patterns mainly utilized by unauthorized sellers. This time, we will look in detail at how those patterns appear in the actual field and how heatmaps objectively expose them.

The strength of heatmaps is that they do not stop at simply drawing the conclusion “same/different,” but perform roles of strengthening the grounds for potential reports (A), clearly distinguishing the scope of reports (B), or preventing false reports (C) depending on the situation.

Situation A: When a seller only changes the background color

: The most common case that appears is the modification of the background color. This involves changing a white background to gray, or applying a warmer or cooler color tone than the original. Although consumers still perceive it as the same product, it is not uncommon for reports to be dismissed during the platform review process solely because “the background has changed.” Indeed, such cases are repeatedly reported even in large marketplaces like Coupang and Naver.

In this case, the heatmap provides a powerful basis for the practitioner. Since the product body (bottle outline, label, lid) is identical, blue is stably displayed on the heatmap, while red tones are concentrated on the altered background area. In other words, the fact that the body is the same and only the background has changed is clearly revealed by the color difference. It objectively proves that even though the background has changed, the essential product image is identical.

Based on this data, practitioners can submit analysis results to the platform stating that “the background has changed, but the core product elements are identical to the original,” rather than simply saying “it is similar.” In internal reporting, “product identity” can also be objectively proven, so it is no longer dismissed as an 'argument based on feel.' Reports that had been frustratingly dismissed now gain clear persuasiveness through heatmaps.

Situation B: When part of the package composition has changed

: Another method is to take an existing product image and cleverly synthesize a part of the package or components. Typical examples include stretching a single product image to make it look like a bundle set, or conversely, exposing it as if it were an authentic product even after removing some products from the original composition. At first glance, it may give the impression of being a “new image,” but in reality, it is an unauthorized use that utilizes the original image as it is.

As an actual case, there was an imitation case registered where the cream container on the far right was deleted from a client company's authentic set image. From the consumer's perspective, it is difficult to notice the difference because the set composition looks like it was originally like this. However, as a result of heatmap analysis, reddish colors were concentrated only in the deleted section, clearly revealing the area where the difference occurred. On the other hand, the product body and the rest of the composition were displayed in deep blue, showing high similarity to the original, thereby confirming that identity was maintained. In other words, it allows you to see at a glance that “the common areas remaining in blue are unauthorized use of the original, and the deleted sections marked in red are modifications.”

Thanks to this, practitioners can obtain visual evidence that “compared to the authentic product, this part has been deleted or modified,” instead of simply claiming that “the composition has changed.” In the internal reporting stage, they can clearly distinguish which part is unauthorized use and which part is a modification, and during platform reporting, they can precisely present the core infringement elements.

Situation C: When it is a completely different product but looks similar

: The key to the previous two situations (A and B) was preparing evidence for cases where our brand's authentic products were cleverly modified and used without authorization. On the other hand, Situation C has a slightly different texture. This time, it is not our brand's product, but it looks similar to consumers. In such cases, there arises a need to prevent false reports instead. It can be confusing for practitioners, but recklessly filing a report with a platform only to have it rejected due to a lack of evidence can also pose a burden on brand trust.

In fact, on Naver Smart Store, there was a case in the cosmetics category where another company's product, photographed in a layout similar to a specific brand's product, was registered, causing consumer confusion. If you place a similar container shape on a white background and even match the label color tone similarly, it can look like a theft at first glance. However, looking at the heatmap analysis, core elements such as the product label or bottle shape are displayed in blue, showing almost no match with the original, and reddish colors appear weakly only in some areas, showing a pattern where differences are dispersed. In other words, the fact that identity is lacking can be confirmed on a color scale. As a result, the similarity score is also calculated as low, clearly proving that “it only looks similar on the surface, but is not actually the same product.”

Thanks to this, practitioners can focus on real cases of unauthorized use instead of wasting resources on unnecessary reports. At the same time, in the internal reporting process, clear explanations like “this looks similar, but based on AI standards, it does not constitute unauthorized use” become possible. In the end, Situation C shows a practical value different from A and B in that it is not about catching everything unconditionally, but catching only what needs to be caught. By preventing false reports, the brand's response strategy becomes more sophisticated, and practitioners are enabled to make “evidence-based selections.”

As such, a heatmap is not just a visualization tool, but functions as a system of evidence, that is, a practical guideline, that guides practitioners on what decisions they can make in what situations. This data carries powerful persuasiveness anywhere: in reports, internal review materials, and platform report documents. When only the background color is different, it highlights identity to increase the likelihood of a successful report (A); when the package composition is partially changed, it reveals commonality and differences simultaneously to clearly define the scope (B); and in the case of completely different products, it confirms distinctness to prevent false reports (C)

. Ultimately, brand managers no longer rely on intuition alone, but can make strategic decisions based on data.

Reatrics provides image similarity diagnostic services, including this heatmap analysis. Ultimately, the key is not simple promotion, but establishing a system where practitioners can convert facts they only knew by intuition into objective evidence. When supported by such a system, brands can protect their assets without being shaken by clever modifications.

A Practical Operation Guide to Integrating Heatmaps into Work

It is also important not to let heatmaps end as simple analysis results, but to actually integrate them into daily work processes. Below is a basic flow that brand practitioners can apply immediately.

First, in the collection phase, secure the URL and screenshots of the suspect product, and record the upload time, seller ID, and option name together. Keeping metadata like this allows for tracking even after a report is made. Next, in the analysis phase, pair the original image with the problematic image, normalize them to the same size and ratio, and then generate a heatmap.

Once the analysis is complete, a review phase is necessary. Here, the key is “where did the heatmap display colors?” The heatmap displays areas with high similarity in blue, and as similarity decreases, it shows them in orange to red. Therefore, checking the color distribution becomes the core criterion for determining whether unauthorized use has occurred.

Thereafter, in the classification phase, divide the cases into the three types described earlier.

• Type A: When only the background or tone is changed

• Type B: When components are deleted/synthesized or banners are added

• Type C: When another company's product is photographed in a similar layout

This classification result leads to evidence packaging. Organize comparative captures placing the original, suspect, and heatmap side-by-side, comment summaries, and logs such as analysis dates and tool versions. Finally, in the action phase, go through the internal approval process and choose one of the following: platform report, seller warning, or hold (to prevent false reporting). Once all processes are completed, conduct a weekly retrospective to summarize the reasons for rejections and update internal guides to reduce repetitive mistakes.

The key in this process is the method of writing comments. Instead of simply saying “it is similar,” you must leave clear sentences explaining “which parts are different” based on the areas highlighted by the heatmap to be persuasive.

• Type A (Background/Tone Change): Product body maintains blue, yellow to red concentrated only on the background area → Confirms the product is identical and only the background has changed

• Type B (Composition Synthesis): Body and label maintain blue, red concentrated on deleted/synthesized sections → Pattern of original theft + partial modification

• Type C (Other Company's Similar Layout): Absence of continuous blue pattern across the body, colors are dispersed, only weak red is formed, no consistent matching pattern → Classified as superficially similar but different products (prevents false reports)

Comments standardized in this way not only serve for internal reporting but also act as powerful evidence in documents submitted to platforms. As such, heatmaps go beyond being a tool for finding stolen images and establish themselves as a device that standardizes practical processes and increases the credibility of reports and filing documents.

Conclusion: How to Overcome the Wall of 'Insufficient Evidence for Reporting'

The biggest repeated frustration in the field of brand protection is that even if there is conviction that “this is indeed unauthorized use,” there is a lack of ways to prove it. Blocked by minor variations, reports are rejected, and unauthorized sellers take advantage of the gap to openly continue selling. This vicious cycle dampens the motivation of practitioners and can create cracks in the brand's trust management system.

However, AI image similarity diagnosis using heatmaps can change this flow. This is because instead of simply saying it is the same or different, it shows visible evidence of where it is the same and where it is different. The fact that the product body remains identical even if the background changes, and the ability to simultaneously reveal differences between synthesized elements and the original, serve as persuasive weapons in both platform and internal reporting.

Ultimately, brand protection is completed not by 'feeling,' but by visual data and quantitative evidence. The moment the conviction felt by practitioners is converted into an objective language, they will no longer be blocked by the wall of “insufficient evidence” and can protect the brand's rights. We hope this article serves as a small starting point for preparing for that change.