Adding Watermark to Images in Computer Vision using Python

Overview

Watermarking is the process of adding a distinctive pattern to an image. It is essential for protecting documents from improper usage or copyright infringement. Digital artefact rights or credits are granted to writers and business owners to prevent unauthorised duplication or copying of their work. Watermarks may produce a distinctive and one-of-a-kind stamp, signature, or trademark.

Introduction

The powerful Python library known as OpenCV allows us to carry out a number of important operations. This library may be used to track camera motions, create picture filters, extract 3D models of any object, and apply watermarks. An OpenCV method called cv2.addWeighted() may be used to add a watermark to an image.

Syntax

Parameters

src1: The initial picture entered. alpha: The initial input image's weight. src2: The picture used as an input. beta: The second input image's weight. gamma: After blending, a scalar value is applied to each pixel.

Let's continue to develop this function for various watermark types on photos using various implementation strategies.

Types of Watermarks

However, before you start watermarking your photos, you must be aware of the many kinds of watermarks and their applications.

Visible Watermarks

As the name suggests, viewers are likely to detect a watermark right away. Basically, the semi-transparent IDs are placed on the original photographs. These identifiers might be text, an address, a URL, a logo, a QR code, or anything else that shows who owns the photos. Despite being semi-transparent, these allow you to view the original images.

The finest watermarks for advertising and copyright protection for images are visible ones. When it comes to spotting picture infringement, they are quite useless. However, they may be easily removed, cropped, or covered with another watermark using the appropriate tools. But every advantage has a price.

Invisible Watermarks

Invisible watermarks can also be referred to as covert or hidden watermarks. As their name implies, they are invisible, so the unaided eye cannot perceive them. To view them, you'll need specific steganography software or hardware.

Several types of watermarks are not detectable.

• Many of these watermarks are translucent and cannot be seen by the human eye, even though they resemble visible watermarks.
• The lowest-order bit of some specific pixels is inverted; however, this only works if the picture doesn't change. But it's too easy to use.
• Spatial watermarking is an additional technique for invisible watermarking. It makes use of a specific colour band that becomes visible when the colours are divided, as in printing.

Text-based Watermark

Let's create a string with the text that will be the picture's watermark. The next step is to customise the typeface. OpenCV comes with several pre-installed font styles. The (255, 255, 255) will cause the watermark text to appear white. Lastly, we can use the putText() function to add the watermark text to the image.

Explanation:

The offered code starts by using OpenCV to read and show an image. Next, it generates a blank picture called watermark_text with the same dimensions as the original image but black (zero) pixel values. The code uses the given font, size, and position to write the text "CLICKS" in white on this blank picture. Then, to produce a watermarked effect, this watermark text is blended with the original picture using the cv2.addWeighted function, which blends the two images with a defined weight (alpha and beta). The outcome is shown with the watermark text somewhat transparently superimposed over the original image. This method makes sure the watermark is noticeable without totally hiding the underlying image.

Output:

Image-based Watermarks

Here are some essential details regarding using OpenCV to apply image-based watermarks on images:

• Watermarks based on photographs are frequently employed to prevent unauthorised use or distribution of digital images.
• An image-based watermark may be applied to an image using OpenCV's robust image-processing capabilities.
• Care should be taken while designing the watermark picture to minimise interference with the underlying image while yet maintaining visibility.
• It is crucial to employ methods that retain the original image's visual integrity while enabling the watermark to be seen when adding one to an image.
• A binary mask that specifies the watermark's location is created, the original and watermarked images are loaded, the watermark and original images are resized to fit the original image, the watermarked image is blended with the original image using the binary mask, and the watermarked image is saved.

Techniques for Adding Watermarks

Digital Watermarking Techniques

Watermarking is the technique of adding side information explicitly specified by the information's owner to the source (cover) data. Digital watermarking hides the owner's visual information inside the original visual data. It is commonly used in applications involving photos and videos.

Conversely, digital watermarking incorporates the mark into the data such that consumers of the source barely notice the change. The secret information is observable even when the watermarked material is heavily altered; it cannot be erased.

There are several types of watermarking, including visible, invisible, and dual. Algorithms for watermarking may be based on a modified or geographical domain. Spatial domain watermarking involves modifying the pixel values of one or more randomly selected sections of photographs. The picture pixels immediately get the raw data. It might be based on using different patches or the least important bit planes. Transform domain watermarking is based on transform coefficient alteration. DCT and DWT are two transform domain methods that are commonly used.

Spatial Domain Techniques

Digital watermarking, often referred to as spatial domain watermarking, is the process of appending watermarked data-side data defined by the owner-on the original (cover) picture or video. Pixels are used in the spatial realm to define a picture. The spatial domain watermarking procedure alters the colour and intensity of a small number of chosen pixels in order to incorporate the watermark. Compared to the transform domain, the spatial domain watermarking is faster and easier to use, but it is more vulnerable to assaults. The spatial domain techniques are simply applied to any image. For spatial domains, the most often used watermarking algorithm is the Least Significant Bit Modification. This method alters the least significant bits (LSB) of the chosen pixels in the picture. It is possible to use a similar method to other LSB bits of the container image.

The watermarked item may appear many times in the container image. It is enough if there is only one watermark left, even if assaults wipe off most of the others. This is a rather simple method. It is resilient to simple procedures like cropping and noise addition. However, lossy compression will make the watermark ineffective. Additionally, a direct assault that changes all of the LSB bits to "1" may be used to erase the watermark from the cover object with no noticeable damage.

After the LSB plane has been eliminated from the watermarked picture, the watermark is retrieved using an exclusive-or operation and the watermark template.

The cover picture can be preprocessed to prevent the concealed information from being taken straight out of the preprocessing. One way to preprocess the cover picture is to break it into blocks. Next, the neighbouring block relationships are altered to encode the embedded watermark.

Frequency Domain Techniques

The fundamental goal of frequency domain watermarking is to include the watermarks in the spectral coefficients of the picture. The three most often used transformations are the Discrete Wavelet Transform (DWT), Discrete Fourier Transform (DFT), and Discrete Cosine Transform (DCT). Discrete cosine transform (DCT) domain watermarking is immune to assaults such as sharpening, filtering, noise, and compression. The standard JPEG compression method serves as its basis, while watermarking is an optional feature. DCT Transform Digital Watermarking alternates the frequency coefficients as opposed to spatial domain watermarking, which changes the pixel LSB of the image bit plane.

Since DCT allows an image to be broken into many frequency bands, it simplifies the process of inserting watermarking data into an image's frequency bands. It is appropriate to embed in the middle-frequency ranges. It strengthens defences against attacks that often can magnify higher frequencies. By adopting intermediate frequencies, the most aesthetically pleasing areas of the image-which are concentrated on the lower frequencies-are likewise avoided. The original image is divided into 8 by 8-pixel blocks, and the DCT transform is applied to each block independently. The middle-frequency range may be extracted from the DCT coefficients by using a middle-frequency 2-D mask. Modern wavelet transform methods are widely applied to watermarking, compression, digital image processing, and other related tasks. The transformations are based on wavelets, which are small waves with a finite period and a variable frequency.

The picture is split into three spatial directions by the wavelet transform: diagonal, horizontal, and vertical. The DWT coefficients have a greater magnitude in the lowest bands (LL) and a lesser magnitude in the other bands (HH, LH, and HL) at every level of decomposition. Wavelets, therefore, better represent the anisotropic properties of HVS.

Many signal processing applications, including the compression of audio and video, the elimination of background noise from audio, and the modelling of wireless antenna layout, commonly use the Discrete Wavelet Transform (DWT).

The discrete Fourier transform (DFT) may split a continuous function into its frequency components. It is impervious to geometrical assaults such as translation, scaling, rotation, and cropping.

The translation invariance is shown using DFT. Spatial changes in the picture do not affect the Fourier transform's amplitude; rather, they change the way the image is characterised in terms of its phase rather than its magnitude.

Transform domain techniques

In the transform domain, a picture is represented in terms of frequencies. Many reversible transformations, such as the discrete cosine transform (DCT), discrete wavelet transform (DWT), discrete Fourier transform (DFT), contourlet transform (CT), and singular value decomposition (SVD), can be used to convert a picture to its frequency representation. Every morph has distinct characteristics and presents the image differently.

DCT Watermarking: This method entails modifying the DCT coefficients in order to embed the watermark inside the DCT domain of the picture.

Wavelet Transform Watermarking: This approach embeds the watermark in the high-frequency subbands by using the wavelet transform to break the image up into different frequency subbands.

Singular Value Decomposition (SVD) Watermarking: This approach uses SVD to break down the image into its singular value matrix and embeds the watermark in the singular values.

Applications of Watermarking in Computer Vision

1. Intellectual property protection: Watermarks may safeguard intellectual property by giving a picture a distinctive identity. By doing this, it may be possible to stop unauthorised usage or copying of the photograph.
2. Security of digital images: Watermarks can be applied to digital photographs as a precautionary measure to stop unauthorised access or use. A watermark, for instance, might show that the image is protected to make it harder for someone to use a picture without permission.
3. Watermarking digital images: Watermarks can confirm the origins of digital images. For example, a watermark may indicate that an image was taken by a certain photographer or company, making it simpler to verify its validity.

Challenges and Limitations of Watermarking

There are a few things to think about, even though it may be a helpful tool for content authentication and copyright protection:

1. Vulnerability to Attacks
   Attacks like removal, modification, and cloning can affect watermarks. Attackers may use a variety of techniques, including filtering, cropping, or scaling, to remove or alter the watermark, which may cause the content's authenticity or copyright protection to be lost.
2. Difficulty in Maintaining Image Quality
   Watermarks embedded in the original image may occasionally compromise its quality. While inserting a digital watermark might result in noise or distortions that lower the overall quality of the image, adding a visual watermark, for example, can obscure important areas of the image.
3. Compatibility With Different Image Formats
   Issues with photo format compatibility may hamper watermarking. Certain image formats-like JPEG-may not support particular watermarking methods, while others can lose the watermark after editing or converting to a different format.

Therefore, while watermarking can benefit content protection and authentication, these drawbacks and challenges need to be considered before designing a watermarking system.

Conclusion

• In computer vision, watermarking is a method used to guarantee the legitimacy and ownership of digital pictures.
• There are many different kinds of watermarking methods, such as visible and invisible watermarking, each with its own advantages and disadvantages.
• Watermarking is useful for a number of purposes, such as digital forensics, copyright protection, and authentication. It may be applied to a variety of materials, including photos, movies, and documents.
• On the other side, watermarking has drawbacks and challenges, including susceptibility to attacks, difficulties maintaining image quality, and problems with compatibility across various image formats.
• Watermarking is a helpful computer vision method for maintaining digital photos, and further study and research in this field can help get over the challenges and restrictions that come with it.