Hi Bojan,

To answer your question, check for these references:

• Confidence interval and confidence level (section 4): https://www.google.fr/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0ahUKEwiJ1LOy95TUAhVLHxoKHTX7B6UQFggyMAA&url=https%3A%2F%2Ficube-publis.unistra.fr%2Fdocs%2F2799%2F7390_32.pdf&usg=AFQjCNGMoSh-_zeeFC0ZyjJJ-vB_UANctQ
• Derivation of confidence intervals: https://www.google.fr/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0ahUKEwikv-G395TUAhXKthoKHdh9BqQQFggwMAA&url=http%3A%2F%2Frepository.upenn.edu%2Fcgi%2Fviewcontent.cgi%3Farticle%3D1208%26context%3Dcis_reports&usg=AFQjCNH8s5WKOxR-0sDyzQAelUSWX23Qgg

Regards

This is an excellent question.   In addition to the very help, incisive answer by @Stéphane Breton, there is a bit more to add.

The problem of deciding on relevant feature in object detection in computer vision using either optical senor arrays in single images or in video frames and infrared sensors, there are three basic forms of features to consider, namely,

geometric shape features (relevant vertices, straight and curved line segments, rectilinear and curvilinear triangles covering neighbourhoods of shapes) ,

shape contour and interior detection (measurement of contour length and measurement of contour interior), and

perceived shape features such as colour, brightness and texture.

A very rich view of relevant object features is given in

https://www.researchgate.net/publication/13194212_Development_of_features_in_object_concepts

In this paper, see Table 1 on page 8.

For a detailed study of object feature detection in video frame analysis, see, e.g.:

https://www.researchgate.net/publication/228811880_A_real-time_system_for_high-level_video_representation_Application_to_video_surveillance

Article Development of features in object concepts

Article A real-time system for high-level video representation: Appl...

My previous post focused on computer stereo-vision. But in a single image feature detector context, I suggest that you check for the following paper by Meer et al., "Edge Detection with Embedded Confidence", 2001 ( http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.21.2946&rep=rep1&type=pdf ).

Miller et al. introduced a confidence measure of interest that they integrated into gradient based edge detectors.

To compute a confidence interval, you first need to determine if your data is continuous or discrete binary. Continuous data are metrics like rating scales, task-time, revenue, weight, height or temperature, etc. Discrete binary data takes only two values, pass/fail, yes/no, agree/disagree and is coded with a 1 (pass) or 0 (fail).

To compute a 95% confidence interval, you need three pieces of data:

The mean (for continuous data) or proportion (for binary data)

The standard deviation, which describes how dispersed the data is around the average

The sample size

Imagine you asked 50 users how satisfied they were with their recent experience with your product on an 7 point scale, with 1 = not at all satisfied and 7 = extremely satisfied.

Find the mean by adding up the scores for each of the 50 users and divide by the total number of responses (which is 50).

For this example, I have an average response of 6.

Compute the standard deviation: You can use the Excel formula = STDEV() for all 50 values or the online calculator. I have a sample standard deviation of 1.2.

Compute the standard error by dividing the standard deviation by the square root of the sample size: 1.2/ √(50) = .17.

Compute the margin of error by multiplying the standard error by 2. 17 x 2 = .34.

Compute the confidence interval by adding the margin of error to the mean from Step 1 and then subtracting the margin of error from the mean:

5.96+.34=6.3

5.96-.34=5.6

We now have a 95% confidence interval of 5.6 to 6.3. Our best estimate of what the entire user population’s average satisfaction is between 5.6 to 6.3.