OE Magazine:<br> <br>Identifying falsified images can be straightforward if you know a few tricks. <br><br>Robert D. Fiete<br><br>Like it or not, fake images are everywhere and have become a part of today's culture. Thanks to the popularity of digital cameras and the availability of desktop imaging software that allows users to easily manipulate images, fake images have become commonplace, especially on the Internet. We see many images that defy common sense and it is natural for us to question the authenticity of these images. Most of have seen images that are obvious fakes, such as the 80-foot grasshopper climbing the Empire State Building, but we naturally assume that these images are fake and know that they were created simply for our amusement. Unfortunately there are too many times when a fake image has been created but it is advertised as real, challenging us to decide for ourselves whether the image is real of fake...<br><br>...In general, fake images are created to generate a deception, but not all fake images are bad. The motivation may be simply for harmless entertainment, which accounts for most fake images generated today. Fake images can be generated for research and development purposes, e.g. to understand image quality issues with different camera designs. The fake images that concern us most are those that are created to perpetuate a lie....Probably the most dangerous motive for generating fake images is to alter the public's perception of truth for political reasons. It would be nice a reliable method existed for determining if an image is real or fake, but unfortunately none exists...<br> <br>Creating Fake Images<br>Although generating fake images historically originated with darkroom tricks, today almost every fake image is made using a computer. Even though it is getting more difficult to discern a real image from a fake image as image processing software improves, image analysis can still be used to detect traits that can expose many of them as fakes...The two most common methods today for generating fake images are to "paint" a new image outright or to alter an existing image that has been captured by a camera...<br><br>Since a digital image is simply a grid of numbers, it is conceivable for an artist to create a computer-generated image by "painting" a grid of numbers to represent any object or scene that could be captured with a digital camera...If the computer generated image is to look like a real photograph, then the image must be consistent with all of the laws of physics applicable to generating a real image.<br><br>Many of the classic painters, such as Leonardo Da Vinci, had an amazing talent to incorporate the proper shading, texture, tone, and color into their paintings that were consistent with the real world thus adding an amazing amount of realism to their work. However, their paintings do not look like modern photographs because they do not contain sufficient detail to match all of the physical properties associated with photographic imaging...<br> <br>In order to create a digital image that looks like a real photograph, the correct brightness values must be determined on a pixel-by-pixel basis to match the physical imaging properties, which could take months to years, depending on the image size, without the aid of computer software to perform the calculation...<br><br>...generating impressive detail in fake images using computer graphics, especially in a movie sequence, is still very difficult due to the complex calculations that need to be performed and the software is not accessible to the average PC user.<br><br>The most common method of generating a fake image, due to its simplicity, is to alter an existing image that was captured by a camera...<br><br>Altering images became routine for many political regimes in the 20th century, especially for propaganda. It was not uncommon for some governments to remove people from historic photographs when these people fell out of favor with the ruling party.<br><br>Today, altering the content of an image does not require dark room tricks but merely a PC with image editing software. Desktop software is readily available and easy to use, allowing anyone to quickly and creatively alter images. The easiest approach is to simply cut a section from one image and embed it into another image (see Figure 6). The desktop software allows the creator to modify the extracted image to the appropriate size and rotation. The software on the market today is so easy to use that that pre-school children have little difficulty creating impressive altered images.<br><br>Identifying Fake Images<br><br>Figure 7: Our perception is the first line of defense at identifying fake images. The cat is obviously too big for this breed of cat and the man would need to lean backwards more to properly hold a cat of this weight.<br><br><br>Figure 8: Our perception can fail to detect a fake image if there is no cause for suspicion, such as when TV Guide used Ann-Margret's body for a picture of Oprah Winfrey.<br><br><br>Figure 9: A computer generated image from the movie Armageddon was circulated on the Internet as an actual image of the Columbia disaster taken from a satellite, even though it has a cartoon look to it.<br> <br>If an image is deemed suspicious, then we can first look for clues by visual inspection and then proceed with scientific inspection if necessary. The first line of defense for detecting a possible fake image is our own perception. We have a keen ability to sense that something is wrong with an image and trusting our common sense works most of the time. If an image looks unbelievable, then it probably is unbelievable and is a fake...<br><br>XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX<br><br>If an image looks real and similar images are easily obtained, then it probably is real since there would be no motive to warrant the time and effort to create the fake image. (hmm...)<br><br>XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX<br><br>...Our perception is very sensitive to subtle details in the composure and texture of objects in an image, especially when viewing images of people...The ability to generate realistic computer generated people is improving dramatically over time as software technology and mathematical models progress.<br><br>A fake image created by altering the context is the hardest to positively identify as fake since the image is real and will pass scientific tests on the validity of the image itself. Most fake UFO images cannot be immediately discounted as fake because they are indeed real photographs of objects that the viewer cannot properly identify, leaving the image subject to interpretation. The key to identifying a fake image when the context is altered is to identify aspects of the image that are inconsistent with the image description, i.e. catch the perpetrator in a lie. For example, the time and date claimed may be inconsistent with the sun's position or the known weather conditions for that date...<br><br>Photographs published in 1932 reportedly showing scenes from WWI dogfights were amazing due to their sharpness and clarity...But the amazing clarity was a clue that the images were probably fake because they appeared too sharp given the relatively long exposures required from cameras at the time and the amount of motion and vibration on the airplane. The images were not proven to be fakes until 1984 when the model airplanes used in the images were discovered.<br><br>When the image content has been altered, we focus on the aspect of the image that makes the image unbelievable. Images that have had their content altered will usually have physical inconsistencies in the image that may be apparent under visual inspection. Unfortunately, these inconsistencies are not always apparent in the image and the image may not be proven to be fake until the original unaltered image is discovered... <br>Understanding the image formation properties of a camera can help us to recognize fake images. The step-by-step physical process of forming an image is called the imaging chain and every image must adhere to the physics of an imaging chain, from the radiometric source to the final image product. Image chain analysis methods are used to examine images for evidence that the laws of physics have been broken. Any inconsistencies found within the image can be an indication the image has been altered.<br><br><br>The physical traits of the image that can be assessed include the illumination conditions, edge sharpness, resolution, tone, relative scale, and noise characteristics. Many of the computer animated scenes created for movies and electronic games do not adhere to the laws of physics, but this is usually intentional to save cost and to make the scenes more entertaining. <br><br>A common inconsistency found when the image content is altered is the mismatch of radiometric or illumination conditions between the altered part and the rest of the image... <br><br>One must be very careful when analyzing the illumination characteristics of the scene. The shadows and illumination conditions can be misleading, especially if the three-dimensional aspects of the scene are not taken into account. The Apollo 11 moon landing images appear to contain "anomalies" that some people use to argue that the moon landing was staged in a studio. These "anomalies" include shadows on the lunar surface that are not parallel and objects that appear illuminated even though they are in the shadows, both suggesting that there were light sources other than the sun, as well as the lack of stars in the black sky, suggesting that a black back-drop was used on a studio set. Of course, all of these so-called anomalies are exactly what we expect to see in the images if we truly understand the imaging conditions on the lunar surface. The shadows are not parallel as seen in the images because the lunar surface is not flat and the objects are not necessarily parallel to one another in height, the shadows are illuminated from the light scattering off of the lunar surface, and the stars do not appear in the images because the camera exposure was set for the brightness of the lunar surface...<br><br>...an understanding of how image processing alters the image characteristics can lead to signs of alteration...<br><br>The Difficulty of Detecting Fake Images<br>Most of the people generating fake images know little or nothing about the physics of the image chain, yet lots of fake images fool us because they seem to have properties that are consistent with real images. How is this possible? Images with altered context are actual images; hence image analysis will not show that the image itself is inconsistent with physics, only that the perpetrator is being untruthful. Images with altered content will usually show signs of alteration if the image is created quickly and carelessly. The anomalies created in an altered image can be reduced by having an understanding of the imaging chain properties and taking the time and effort to ensure that the entire image is consistent at the pixel level, but this is rarely performed <br><br>XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX<br><br>due to the knowledge and time required.<br><br>XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX<br><br>The simplest method to reduce the detection of the anomalies in an altered image is to degrade the quality of the image of the alteration. The most common methods are blurring the edges, adding random noise, reducing the size of the image, or compressing the image, all of which will cover up telltale signs of the manipulation. Many fake images have such poor quality that accurate measurements cannot be made to determine if inconsistencies exist. Admittedly, most creators of fake images do not reduce the quality with the intent of making image analysis more difficult, but instead reduce the quality by resizing and compressing the image simply to reduce the file size. However, reducing the image quality to hide the inconsistencies may reduce the impact that the creator of the altered image had hoped for...<br><br><br>XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX<br><br>Although image analysis tools can help detect many fake images, currently there is no way to stop somebody from spending the time and resources to make a fake image that is not detectible. <br><br>XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX<br><br>All one can do is hope that an inconsistency can be found, thus indicating that the image is fake... <br><br>Robert Fiete is chief technologist at the space systems division of ITT Industries, Rochester, NY. Contact: 585-253-5772;
robert.fiete@itt.com. <br><br> <!--EZCODE AUTOLINK START--><a href="http://oemagazine.com/fromTheMagazine/jan05/photofakery.html">oemagazine.com/fromTheMag...akery.html</a><!--EZCODE AUTOLINK END--><br> <p></p><i></i>
by proldic » Sun Aug 21, 2005 6:03 pm 
