I am generating images (thumbnails) from a video every 3 seconds. Now I need to discard/remove all the similar images. Is there a way I could this?
I generate thumbnails using FFMPEG. I read about various image-diff solutions like given in this SO post, but I do not want to do this manually. How and what parameters should be considered that could tell if a particular image is similar to other images present.
You can calculate the Structural Similarity Index between images and based on the score keep or discard an image. There are other measures you can use, but basically a method that returns a score. Try PIL or OpenCV
https://pillow.readthedocs.io/en/3.1.x/reference/ImageChops.html?highlight=difference
https://www.pyimagesearch.com/2017/06/19/image-difference-with-opencv-and-python/
I dont have enough reputation to comment my idea on your problem, so i will just go ahead and post it as an answer in hope of helping you.
I am quite confused about the term "similar" but since you are reffering on video frames i am going to assume that you want to avoid having "similar" frames that have been captured because of poor camera movement. If that's the case you might want to consider using salient point descriptors.
To be more specific you can detect salient points (using for instance Harris) and then use a point descriptor algorithm (such as SURF) and discard the frames that have been found to have "too many" similar points with a pre-selected frame.
Keep in mind that in order for the above process to be successful, the frames must be as sharp as possible, i guess you don't want to extract as a thubnail a blurred frame anyway. So applying a blurred images detection might be useful in your case.
Related
I'm currently an intern at a quality inspector company. My job is to write a program that can detect faulty products (for example, missing screw). They take a picture of every single product. My idea is that I choose an image which could serve as a benchmark and I would compare the other images to that, with the SSIM score, and maybe display the faulty part with a rectangle. Is this a viable idea? (Its a strange internship, because it seems like I'm the only one who can code there...) that's why I'm asking here.
It sounds good idea if your goal is to classify different objects within images comparing benchmark image.
But in my experience, SSIM score was sensitive to angle, light or environment.
So in conclusion, if your goal is to classify different objects in images, your idea would work. But if your goal is to classify exactly same objects, it might not be able to classify.
I have been going over this for days now and have hit a road block as I am too scared to try out my hypothesis.
I would like to find out the number of grayed rectangular boxes in this image. However, I am not sure how I can do that. I was thinking of two ways:
i. Getting area of the connected components, calculating their median and getting the number of components between a certain percentile of the area (may sound pretty strange).
ii. Making a machine learning model and find out the similar boxes in the image and count them.
However, I would like them to be more generalized so that I will need to be able to make the solution fit other images that I would need to be processed.
Here is my source Image:
Any sort of help/suggestions and even solutions would be greatly appreciated.
Thanks in advance!
Maybe you are losing a lot of image information with filtering...Do you have an unfiltered source image too? I suppose ML approach would work pretty nice then.
I noticed you could achieve better resolution if your camera is 90 rotated (If you could affect this)
for my school project, I need to find images in a large dataset. I'm working with python and opencv. Until now, I've managed to find an exact match of an image in the dataset but it takes a lot of time even though I had 20 images for the test code. So, I've searched few pages of google and I've tried the code on these pages
image hashing
building an image hashing search engine
feature matching
Also, I've been thinking to search through the hashed dataset, save their paths, then find the best feature matching image among them. But most of the time, my narrowed down working area is so much different than what is my query image.
The image hashing is really great. It looks like what I need but there is a problem: I need to find an exact match, not similar photos. So, I'm asking you guys, if you have any suggestion or a piece of code might help or improve the reference code that I've linked, can you share it with me? I'd be really happy to try or research what you guys send or suggest.
opencv is probably the wrong tool for this. The algorithms there are geared towards finding similar matches, not exact ones. The general idea is to use machine learning to teach the code to recognize what a car looks like so it can detect cars in videos, even when the color or form changes (driving in the shadow, different make, etc).
I've found two approaches work well when trying to build an image database.
Use a normal hash algorithm like SHA-256 plus maybe some metadata (file or image size) to find matches
Resize the image down to 4x4 or even 2x2. Use the pixel RGB values as "hash".
The first approach is to reduce the image to a number. You can then put the number in a look up table. When searching for the image, apply the same hashing algorithm to the image you're looking for. Use the new number to look in the table. If it's there, you have a match.
Note: In all cases, hashing can produce the same number for different pictures. So you have to compare all the pixels of two pictures to make sure it's really an exact match. That's why it sometimes helps to add information like the picture size (in pixels, not file size in bytes).
The second approach allows to find pictures which very similar to the eye but in fact slightly different. Imagine cropping off a single pixel column on the left or tilting the image by 0.01°. To you, the image will be the same but for a computer, they will by totally different. The second approach tries to average small changes out. The cost here is that you will get more collisions, especially for B&W pictures.
Finding exact image matches using hash functions can be done with the undouble library (Disclaimer: I am also the author). It works using a multi-step process of pre-processing the images (grayscaling, normalizing, and scaling), computing the image hash, and the grouping of images based on a threshold value.
How do you detect the location of an image within a larger image? I have an unmodified copy of the image. This image is then changed to an arbitrary resolution and placed randomly within a much larger image which is of an arbitrary size. No other transformations are conducted on the resulting image. Python code would be ideal, and it would probably require libgd. If you know of a good approach to this problem you'll get a +1.
There is a quick and dirty solution, and that's simply sliding a window over the target image and computing some measure of similarity at each location, then picking the location with the highest similarity. Then you compare the similarity to a threshold, if the score is above the threshold, you conclude the image is there and that's the location; if the score is below the threshold, then the image isn't there.
As a similarity measure, you can use normalized correlation or sum of squared differences (aka L2 norm). As people mentioned, this will not deal with scale changes. So you also rescale your original image multiple times and repeat the process above with each scaled version. Depending on the size of your input image and the range of possible scales, this may be good enough, and it's easy to implement.
A proper solution is to use affine invariants. Try looking up "wide-baseline stereo matching", people looked at that problem in that context. The methods that are used are generally something like this:
Preprocessing of the original image
Run an "interest point detector". This will find a few points in the image which are easily localizable, e.g. corners. There are many detectors, a detector called "harris-affine" works well and is pretty popular (so implementations probably exist). Another option is to use the Difference-of-Gaussians (DoG) detector, it was developed for SIFT and works well too.
At each interest point, extract a small sub-image (e.g. 30x30 pixels)
For each sub-image, compute a "descriptor", some representation of the image content in that window. Again, many descriptors exist. Things to look at are how well the descriptor describes the image content (you want two descriptors to match only if they are similar) and how invariant it is (you want it to be the same even after scaling). In your case, I'd recommend using SIFT. It is not as invariant as some other descriptors, but can cope with scale well, and in your case scale is the only thing that changes.
At the end of this stage, you will have a set of descriptors.
Testing (with the new test image).
First, you run the same interest point detector as in step 1 and get a set of interest points. You compute the same descriptor for each point, as above. Now you have a set of descriptors for the target image as well.
Next, you look for matches. Ideally, to each descriptor from your original image, there will be some pretty similar descriptor in the target image. (Since the target image is larger, there will also be "leftover" descriptors, i.e. points that don't correspond to anything in the original image.) So if enough of the original descriptors match with enough similarity, then you know the target is there. Moreover, since the descriptors are location-specific, you will also know where in the target image the original image is.
You probably want cross-correlation. (Autocorrelation is correlating a signal with itself; cross correlating is correlating two different signals.)
What correlation does for you, over simply checking for exact matches, is that it will tell you where the best matches are, and how good they are. Flip side is that, for a 2-D picture, it's something like O(N^3), and it's not that simple an algorithm. But it's magic once you get it to work.
EDIT: Aargh, you specified an arbitrary resize. That's going to break any correlation-based algorithm. Sorry, you're outside my experience now and SO won't let me delete this answer.
http://en.wikipedia.org/wiki/Autocorrelation is my first instinct.
Take a look at Scale-Invariant Feature Transforms; there are many different flavors that may be more or less tailored to the type of images you happen to be working with.
I have a camera that will be stationary, pointed at an indoors area. People will walk past the camera, within about 5 meters of it. Using OpenCV, I want to detect individuals walking past - my ideal return is an array of detected individuals, with bounding rectangles.
I've looked at several of the built-in samples:
None of the Python samples really apply
The C blob tracking sample looks promising, but doesn't accept live video, which makes testing difficult. It's also the most complicated of the samples, making extracting the relevant knowledge and converting it to the Python API problematic.
The C 'motempl' sample also looks promising, in that it calculates a silhouette from subsequent video frames. Presumably I could then use that to find strongly connected components and extract individual blobs and their bounding boxes - but I'm still left trying to figure out a way to identify blobs found in subsequent frames as the same blob.
Is anyone able to provide guidance or samples for doing this - preferably in Python?
The latest SVN version of OpenCV contains an (undocumented) implementation of HOG-based pedestrian detection. It even comes with a pre-trained detector and a python wrapper. The basic usage is as follows:
from cv import *
storage = CreateMemStorage(0)
img = LoadImage(file) # or read from camera
found = list(HOGDetectMultiScale(img, storage, win_stride=(8,8),
padding=(32,32), scale=1.05, group_threshold=2))
So instead of tracking, you might just run the detector in each frame and use its output directly.
See src/cvaux/cvhog.cpp for the implementation and samples/python/peopledetect.py for a more complete python example (both in the OpenCV sources).
Nick,
What you are looking for is not people detection, but motion detection. If you tell us a lot more about what you are trying to solve/do, we can answer better.
Anyway, there are many ways to do motion detection depending on what you are going to do with the results. Simplest one would be differencing followed by thresholding while a complex one could be proper background modeling -> foreground subtraction -> morphological ops -> connected component analysis, followed by blob analysis if required. Download the opencv code and look in samples directory. You might see what you are looking for. Also, there is an Oreilly book on OCV.
Hope this helps,
Nand
This is clearly a non-trivial task. You'll have to look into scientific publications for inspiration (Google Scholar is your friend here). Here's a paper about human detection and tracking: Human tracking by fast mean shift mode seeking
This is similar to a project we did as part of a Computer Vision course, and I can tell you right now that it is a hard problem to get right.
You could use foreground/background segmentation, find all blobs and then decide that they are a person. The problem is that it will not work very well since people tend to go together, go past each other and so on, so a blob might very well consist of two persons and then you will see that blob splitting and merging as they walk along.
You will need some method of discriminating between multiple persons in one blob. This is not a problem I expect anyone being able to answer in a single SO-post.
My advice is to dive into the available research and see if you can find anything there. The problem is not unsolvavble considering that there exists products which do this: Autoliv has a product to detect pedestrians using an IR-camera on a car, and I have seen other products which deal with counting customers entering and exiting stores.