How can photos be manipulated

Most photo editing software comes with some form of Healing Brush , Clone and Stamp , and Transform tools. These are all the bread and butter of photo manipulation. Graphic tablets are good for more than just drawing or painting on a digital canvas. The Wacom Intuos is an excellent, affordable option to get you started. The Wacom Cintiq 16 is also a popular option, perfect for regular Photoshop use. When you come back with fresh eyes, do one final check. Take a look at the shadows and highlights, light direction, and any areas where you added or subtracted something.

Adobe Photoshop is the original layer-based non-destructive editing program, although many other editing apps also offer it these days. As mentioned before, what software you choose to use will make all the difference in the world. Ideally, your program will have adjustment layers , excellent selection tools , and a healing brush , clone stamp , and refine edge tool that work really , really well. Desktop computers are better value for money than laptops and traditionally offer better performance too, although most modern laptops can handle Photoshop and other editing apps with no issues.

Like all skills, effective photo manipulation takes practice. If you have Adobe Photoshop and are yearning to get your feet wet in the art of photo manipulation, check out these amazing tutorials for photographers and graphic designers. They all come with stock images and clear, concise directions for each and every step.

These Photoshop tutorials are a bit involved, however, so make sure you have a chunk of time set aside to play around with. Ever wish you could make your toys come to life? This tutorial covers not only how to manipulate images, but how to create a photoshoot for the different elements as well. Get your camera ready! It shows you how to create a symmetrical background, add in the reflections, and warp a face to make it look like that of a child. You might also find some inspiration in our guide to surreal photography.

Ever wonder what a giraffe looks like without its spots? This tutorial will take you through not only removing its spots, but also turning them into an outfit to be ironed. This tutorial will show you how to seamlessly blend all the stock photos together into soft, autumnal, whole.

Need some more inspiration? We distinguished between physically implausible versus plausible manipulations. Such shadows imply the impossible: two suns. Alternatively, when an unfamiliar face is retouched in an image it is quite plausible; eliminating spots and wrinkles or whitening teeth do not contradict physical constraints in the world that govern how faces ought to look.

In our study, geometrical and shadow manipulations made up our implausible manipulation category, while airbrushing and addition or subtraction manipulations made up our plausible manipulation category. Our fifth manipulation type, super-additive , presented all four manipulation types in a single image and thus included both categories of manipulation. In particular, people should correctly identify more of the physically implausible manipulations than the physically plausible manipulations given the availability of evidence within the photo.

We also expected people to be better at correctly detecting and locating manipulations that caused more change to the pixels in the photo than manipulations that caused less change. A further 17 subjects were excluded from the analyses because they had missing response time data for at least one response on the detection or location task.

There were no geographical restrictions and subjects did not receive payment for taking part, but they did receive feedback on their performance at the end of the task. Subject recruitment stopped when we reached at least responses per photo. We used a within-subjects design in which each person viewed a series of ten photos, half of which had one of five manipulation types applied, and half of which were original, non-manipulated photos. The first author SN used GNU Image Manipulation Program GIMP to apply five different, commonly used manipulation techniques: a airbrushing, b addition or subtraction, c geometrical inconsistency, d shadow inconsistency, and e super-additive manipulations a to d included within a single image.

For the addition or subtraction technique, we added or removed objects, or parts of objects. For example, we removed links between tower columns on a suspension bridge and inserted a boat into a river scene. For geometrical inconsistencies, we created physically implausible changes, such as distorting angles of buildings or sheering trees in different directions to others to indicate inconsistent wind direction.

For shadow inconsistencies, we removed or changed the direction of a shadow to make it incompatible with the remaining shadows in the scene.

In the super-additive technique we presented all four previously described manipulation types in one photo. Figure 1 shows examples of the five manipulation types, and higher resolution versions of these images, as well as other stimuli examples, appear in Additional file 1.

Samples of manipulated photos. In total, we had ten photos of different real-world scenes. The non-manipulated version of each of these ten photos was used to create our original photo set. To generate the manipulated photos, we applied each of the five manipulation types to six of the ten photos, creating six versions of each manipulation for a total of 30 manipulated photos. This gave us an overall set of 40 photos.

Subjects saw each of the five manipulation types and five original images but always on a different photo. Image-based saliency cues can determine where subjects direct their attention; thus, we checked whether our manipulations had changed the salience of the manipulated area within the image.

To summarize, we found that our manipulations did not inadvertently change the salience of the manipulated regions. See Additional file 2 for details of these analyses. Subjects answered questions about their demographics, attitudes towards image manipulation, and experiences of taking and manipulating photos.

Subjects were then shown a practice photo and instructed to adjust their browser zoom level so that the full image was visible. Next, subjects were presented with ten photos in a random order and they had an unlimited amount of time to view and respond to each photo.

For the analyses we considered a response to be correct if the subject clicked on a region that contained any of the manipulated area or a nearby area that could be used as evidence that a manipulation had taken place—a relatively liberal criterion.

Subjects received feedback on their performance at the end of the study. An analysis of the response time data suggested that subjects were engaged with the task and spent a reasonable amount of time determining which photos were authentic.

In the detection task, the mean response time per photo was In the location task, the mean response time was We now turn to our primary research question: To what extent can people detect and locate manipulations of real-world photos? Furthermore, even when subjects correctly indicated that a photo had been manipulated, they could not necessarily locate the manipulation.

To determine chance performance in the location task, we need to take into account that subjects were asked to select one of nine regions of the image. Therefore, subjects had less chance of being correct by guessing in the location task than the detection task. On average, the manipulations were contained within two of the nine regions. But because the chance of being correct by guessing varied for each image and each manipulation type, we ran a Monte Carlo simulation to determine the chance rate of selecting the correct region.

Table 1 shows the results from one million simulated responses. Overall, the results show that people have some above chance ability to detect and locate manipulations, although performance is far from perfect. In line with our prediction, subjects were better at detecting manipulations that included physically implausible changes geometrical inconsistencies, shadow inconsistencies, and super-additive manipulations than images that included physically plausible changes airbrushing alterations and addition or subtraction of objects.

The dotted line represents chance performance for detection. The grey dotted lines on the locate bars represent chance performance by manipulation type in the location task. It was not the case, however, that subjects were necessarily better at locating the manipulation within the photo when the change was physically implausible.

Figure 4 shows the proportion of manipulated photo trials in which subjects correctly detected a manipulation and also went on to correctly locate that manipulation, by manipulation type.

Across both physically implausible and physically plausible manipulation types, subjects often correctly indicated that photos were manipulated but failed to then accurately locate the manipulation. Furthermore, although the physically implausible geometrical inconsistencies were more often correctly located, the shadow inconsistencies were only located equally as often as the physically plausible manipulation types—airbrushing and addition or subtraction.

These findings suggest that people may find it easier to detect physically implausible, rather than plausible, manipulations, but this is not the case when it comes to locating the manipulation. The grey dotted lines on the bars represent chance performance for each manipulation type. When an image is digitally altered, the structure of the underlying elements—the pixels—are changed.

This change can be quantified in numerous ways but we chose to use Delta-E 76 because it is a measure based on both color and luminance Robertson, Next we calculated the difference between corresponding pixels in the original and manipulated versions of each photo.

Finally, these differences were averaged to give a single Delta-E score for each manipulated photo. A higher Delta-E value indicates a greater amount of difference between the original and the manipulated photo.

Footnote 3 We calculated Delta-E for each of the 30 manipulated photos. Figure 5 shows the log Delta-E values on the x-axis, where larger values indicate more change in the color and luminance values of pixels in the manipulated photos compared with their original counterpart. The proportions of correct detection Fig. As predicted, these data suggest that people might be sensitive to the low level properties of real-world scenes when making judgments about the authenticity of photos.

This finding is especially remarkable given that our subjects never saw the same scene more than once and so never saw the original version of a manipulated image. Presumably, these disruptions make it easier for people to accurately classify manipulated photos as being manipulated. Mean proportion of correctly detected a and located b image manipulations by extent of pixel distortion as measured by Delta-E. The graphs show individual data points for each of the 30 manipulated images. Next, we tested whether there was a relationship between the mean amount of change and the mean proportion of correct detection Fig.

As Fig. The graphs show the mean values for each of the five categories of manipulation type. For the detection task, we ran two additional repeated measures linear regression GEE models to explore the effect of the predictor variables on signal detection estimates d' and c. The results of the GEE analyses are shown in Table 2. In the detection task, faster responses were more likely to be associated with accurate responses than slower responses.

Those who believe a greater percentage of photos are digitally manipulated were more likely to correctly identify manipulated photos than those who believe a lower percentage of photos are digitally manipulated. Further, the results of the signal detection analysis suggest that this results from a difference in ability to discriminate between original and manipulated photos, rather than a shift in response bias—those who believe a greater percentage of photos are digitally manipulated accurately identified more of the manipulated photos without an increased false alarm rate.

This pattern of results is somewhat surprising. It seems intuitive to think that a general belief that manipulated photos are prevalent simply makes people more likely to report that a photo is manipulated because they are generally skeptical about the veracity of photos rather than because they are better at spotting fakes. Although interesting, the small effect size and counterintuitive nature of the finding indicate that it is important to replicate the result prior to drawing any strong conclusions.

The only variable that had an effect on accuracy in the location task was gender; males were slightly more likely than females to correctly locate the manipulation within the photo. Together these findings show that individual factors have relatively little impact on the ability to detect and locate manipulations. In fact, our response time findings might be explained by a number of perceptual decision making models, for example, the drift diffusion model Ratcliff, However, determining the precise mechanism that accounts for the association between shorter response times and greater accuracy is beyond the scope of the current paper.

Experiment 1 indicates that people have some ability to distinguish between original and manipulated real-world photos. Our data also suggest that locating photo manipulations is a difficult task, even when people correctly indicate that a photo is manipulated. Recall that subjects were only asked to locate manipulations on photos that they thought were manipulated.

It remains possible people might be able to locate manipulations even if they do not initially think that a photo has been manipulated. We were unable to check this possibility in Experiment 1, so we addressed this issue in Experiment 2 by asking subjects to complete the location task for all photos, regardless of their initial response in the detection task.

If subjects did not think that the photo had been manipulated, we asked them to make a guess about which area of the image might have been changed.

We also created a new set of photographic stimuli for Experiment 2. Rather than sourcing photos online, the first author captured a unique set of photos on a Nikon D40 camera in RAW format, and prior to any digital editing, converted the files to PNGs. There are two crucial benefits to using original photos rather than downloading photos from the web.

First, by using original photos we could be certain that our images had not been previously manipulated in any way.

Second, when digital images are saved, the data are compressed to reduce the file size. JPEG compression is lossy in that some information is discarded to reduce file size. This information is not generally noticeable to the human eye except at very high compression rates when compression artifacts can occur ; however, the process of converting RAW files to PNGs a lossless format prevented any loss of data in either the original or manipulated images and, again, ensured that our photos were not manipulated in any way before we intentionally manipulated them.

A further 32 subjects were excluded from the analyses because they had missing response time data for at least one response on the detection or location task. As in Experiment 1, subjects did not receive payment for taking part but were given feedback on their performance at the end of the study.

We stopped collecting data once we reached responses per photo. The design was similar to that of Experiment 1. We checked the photos to ensure there were no spatial distortions caused by the lens, such as barrel or pincushion distortion. The photo manipulation process was the same as in Experiment 1. We applied the five manipulation techniques to six different photos to create a total of 30 manipulated photos.

We used the non-manipulated version of these six photos and another four non-manipulated photos to give a total of ten original photos. Thus, the total number of photos was As in Experiment 1, we ran two independent saliency models to check whether our manipulations had influenced the salience of the region where the manipulation had been made.

See Additional file 2 for details of the saliency analyses. Similar to Experiment 1, our manipulations made little difference to the salience of the regions of the image. The procedure was similar to that used in Experiment 1, except for the following two changes.

First, subjects were asked to locate the manipulation regardless of their response in the detection task. Second, subjects were asked to click on one of 12, rather than nine, regions on the photo to locate the manipulation. We increased the number of regions on the grid to ensure that the manipulations in the photos spanned two regions, on average, as per Experiment 1.

As in Experiment 1, subjects spent a reasonable amount of time examining the photos. It is possible that asking all subjects to search for evidence of a manipulation—the location task—regardless of their answer in the detection task, prompted a more careful consideration of the scene.

In line with this account, subjects in Experiment 2 spent a mean of 14 s longer per photo on the detection task than those in Experiment 1. Recall that the results from Experiment 1 suggested that subjects found the location task difficult, even when they correctly detected the photo as manipulated. Yet, we were unable to conclusively say that location was more difficult than detection because we did not have location data for the manipulated photo trials that subjects failed to detect.

For the location task, however, there were two differences to Experiment 1. First, subjects were asked to select one of 12, rather than one of nine, image regions. Second, we used a new image set; thus, the number of regions manipulated for each image and manipulation type changed.

Accordingly, we ran a separate Monte Carlo simulation to determine the chance rate of selecting the correct region. This finding suggests that people are better at the more direct task of locating manipulations than the more generic one of detecting if a photo has been manipulated or not. One possibility is that our assumption that each of the 12 image regions has an equal chance of being picked is too simplistic—perhaps certain image regions never get picked e. To check this possibility, we ran a second chance performance calculation.

In Experiment 2, even when subjects did not think that the image had been manipulated, they still attempted to guess the region that had been changed. Therefore, we can use these localization decisions in the original non-manipulated versions of the six critical photos to determine chance performance in the task. This analysis allows us to calculate chance based on the regions of non-manipulated images that people actually selected when guessing rather than assuming each of the 12 regions has an equal chance of being picked.

This finding supports the idea that subjects are better at the more direct task of locating manipulations than detecting whether a photo has been manipulated or not. On the manipulated photo trials, asking subjects to locate the manipulation regardless of whether they correctly detected it allowed us to segment accuracy in the following ways: i accurately detected and accurately located hereafter, DL , ii accurately detected but not accurately located DnL , iii inaccurately detected but accurately located nDL , or iv inaccurately detected and inaccurately located nDnL.

Intuitively, it seems most practical to consider the more conservative accuracy—DL—as correct, especially in certain contexts, such as the legal domain, where it is crucial to know not only that an image has been manipulated, but precisely what about it is fake.

That said, it might be possible to learn from the DnL and nDL cases to try to better understand how people process manipulated images. The most common outcomes were for subjects to both accurately detect and accurately locate manipulations, or both inaccurately detect and inaccurately locate manipulations. Subjects infrequently managed to detect and locate airbrushing manipulations; in fact it was more likely that subjects made DnL or nDL responses.

Although this fits with our prediction that plausible manipulations would be more difficult to identify than implausible ones, the pattern of results for geometrical inconsistency, shadow inconsistency, and addition or subtraction do not support our prediction. Subjects made more DL responses on the plausible addition or subtraction manipulation photos than on either of the implausible types, geometrical manipulations and shadow manipulations.

Why, then, are subjects performing better than expected by either of the chance measures on the addition or subtraction manipulations and worse than expected on the airbrushing ones?

Mean proportion of manipulated photos accurately detected and accurately located DL , accurately detected, inaccurately located DnL , inaccurately detected, accurately located nDL , and inaccurately detected, inaccurately located nDnL by manipulation type. The dotted horizontal lines on the bars represent chance performance for each manipulation type from the results of the Monte Carlo simulation.

Recall that the results from Experiment 1 suggested a relationship between the correct detection and location of image manipulations and the amount of disruption the manipulations had caused to the underlying structure of the pixels. Yet, the JPEG format of the images used in Experiment 1 created some re-compression noise in the Delta-E measurements between different images; thus, we wanted to test whether the same finding held with the lossless image format used in Experiment 2.

As shown in Fig. These Pearson correlation coefficients are larger than those in Experiment 1 cf. It is possible that the re-compression noise in the JPEG images in Experiment 1 obscured the relationship between Delta-E and detection and localization performance. This finding suggests that Delta-E is a more useful measure for local, discrete changes to an image than it is for global image changes, such as applying a filter.

Of course, the whole point of manipulating images is to fool observers, to make them believe that something fake is in fact true. Therefore, it might not be particularly surprising to learn that people find it difficult to spot high quality image manipulations. Yet it is surprising to learn that, even though our subjects never saw the same image more than once, this ability might be dependent on the amount of disruption between the original and manipulated image.

Our findings suggest that manipulation type and the technique used to create the manipulation, for instance, cloning or scaling, might be less important than the extent to which the change affects the underlying pixel structure of the image.

To test this possibility, we next consider the relationship between the Delta-E values and the proportion of a correct detection and b location responses by the category of manipulation type. That is, subjects accurately detected and located more of the addition or subtraction manipulations than the geometry, shadow, or airbrushing manipulations. One possibility is that the five categories of manipulation type introduced different amounts of change between the original and manipulated versions of the images.

To check this, we calculated the mean proportion of correct detections, localizations, and Delta-E values for each of the five categories of manipulation type. These results suggest that the differences in detection and localization rates across the five manipulation types are better accounted for by the extent of the physical change to the image caused by the manipulation, rather than the plausibility of that manipulation.

Yet, given that subjects did not have the opportunity to compare the manipulated and original version of the scene, it is not entirely obvious why amount of change predicts accuracy. Our results suggest that the amount of change between the original and manipulated versions of an image is an important factor in explaining the detectability and localization of manipulations.

Next we considered whether any individual factors are associated with improved ability to detect or locate manipulations. As discussed, we were able to use liberal or stringent criteria for our classification of detection and location accuracy on the manipulated image trials.

Accordingly, we ran three models: the first two used the liberal classification for accuracy and replicated the models we ran in Experiment 1 , and the other examined the more stringent classification, DL.

As in Experiment 1, for the detection task, we also ran two repeated measures linear regression GEE models to explore the effect of the predictor variables on signal-detection estimates d' and c.

We included the same factors used in the GEE models in Experiment 1. The results of the GEE analyses are shown in Table 5. Using the more liberal accuracy classification, that is, both DL and DnL responses for detection, we found that three factors had an effect on likelihood to respond correctly: response time, general beliefs about the prevalence of photo manipulation, and interest in photography.

As in Experiment 1, faster responses were more likely to be correct than slower responses. Also replicating the finding in Experiment 1, those who believe a greater percentage of photos are digitally manipulated were slightly more likely to correctly identify manipulated photos than those who believe a lower percentage of photos are digitally manipulated.

Additionally, in Experiment 2, those interested in photography were slightly more likely to identify image manipulations correctly than those who are not interested in photography. For the location task, using the more liberal accuracy classification, that is, both DL and nDL responses, we found that two factors had an effect on likelihood to respond correctly.

Again there was an effect of response time: In the location task, faster responses were more likely to be correct than slower responses. Also those with an interest in photography were slightly more likely to correctly locate the manipulation within the photo than those without an interest. Next we considered whether any factors affected our more stringent accuracy classification, that is, being correct on both the detection and location tasks DL.

The results revealed an effect for two factors on likelihood to respond correctly. Specifically, there was an effect of response time with shorter response times being associated with greater accuracy. There was also an effect of interest in photography, with those interested more likely to correctly make DL responses than those not interested.

Our GEE models in both Experiments 1 and 2 revealed that shorter response times were linked with more correct responses on both tasks. As in Experiment 1, this association might be explained by several models of perceptual decision making; however, determining which of these models best accounts for our data is beyond the scope of the current paper. Considering the prevalence of manipulated images in the media, on social networking sites, and in other domains, our findings warrant concern about the extent to which people may be frequently fooled in their daily lives.

Furthermore, we did not find any strong evidence to suggest that individual factors, such as having an interest in photography or beliefs about the extent of image manipulation in society, are associated with improved ability to detect or locate manipulations. Recall that we looked at two categories of manipulations—implausible and plausible—and we predicted that people would perform better on implausible manipulations because these scenes provide additional evidence that people can use to determine if a photo has been manipulated.

Yet the story was not so simple. In Experiment 1, subjects correctly detected more of the implausible photo manipulations than the plausible photo manipulations, but in Experiment 2, the opposite was true. Further, even when subjects correctly identified the implausible photo manipulations, they did not necessarily go on to accurately locate the manipulation. Placing disclaimers on photos stating that they have been manipulated does not reduce their effects on body dissatisfaction, as women still find photos realistic even when told they were digitally manipulated.

As Dr. In fact, young girls often use photo manipulation software to retouch their own photos. Blake Lively says AI Standard. Emily Ratajowski slams her own cover on Instagram.

CBC News. American Medical Association, June 21, Boston University School of Law. Peter Lang Publishing, New York. Skip to main content.