Method and system for video and film recommendation
An artificial intelligence video analysis recommendation system and method is provided that allows video viewers to discover new videos and video producers to evaluate the potential success of a new project as well as to understand the perceptual factors beneath audience ratings. The system and method accomplish these tasks by analyzing a database of video in order to identify key similarities between different pieces of video, and then recommends pieces of video to a user depending upon their video preferences. An embodiment enables a user to evaluate a new video's similarity to videos already established as commercially valuable.
1. A method of analyzing videos, said method comprising:
providing a digital database comprising a plurality of digital video files;
selecting one of said digital video files for analysis;
if necessary, transforming said one of said digital video files into a selected format comprising sequences of frames or photograms and audio files;
processing said one of said digital video files to extract video descriptors and metadata and to establish a plurality of parameters, said parameters being representative of characteristics of said one of said digital video files,
said automatic perceptual analysis processing further comprising:
processing said one of said digital video files to calculate statistics for parameters selected from the group consisting of:
amount of red,
amount of green,
amount of blue,
saturation,
light level/luminosity,
tone,
light level period, and
Hu's Momentum;
calculating tone and saturation histogram;
evaluating structures made up of lines to determine visual complexity of an image;
calculating an amount of movement between two consecutive frames/photograms;
analyzing scene frequency in grey scale, and for components in red, green, and blue; and
measuring scene complexity using color segmentation;
processing said one of said digital video files to establish correlations among said video descriptors; and
establishing an output measure that relates said video descriptors with audience data.
2. The method according to claim 1 , said processing said one of said digital video files to extract video descriptors and metadata further comprising:
using supervised perceptual analysis to extract numbers of characters and identification of said characters in a scene without using script information.
3. The method according to claim 1 , said processing said one of said digital video files to extract video descriptors and metadata further comprising:
using supervised cognitive analysis to evaluate scene structure and classification for each scene including scene character, scene intensity, and scene type.
4. The method according to claim 1 , said processing said one of said digital video files to extract video descriptors and metadata further comprising:
using automatic perceptual analysis to establish said plurality of parameters.
5. The method according to claim 4 , said automatic perceptual analysis further comprising:
analyzing audio parameters selected from the group consisting of:
sound power,
spectrum fractiles from low frequencies,
maximum frequency and frequency centroid,
power of low, medium, and high frequencies,
power break, and
harmonic detector.
6. The method according to claim 1 , said output measure that relates said video descriptors with audience data being determined for each scene type.
7. The method according to claim 1 , said method measuring innovation cycles or prediction cycles in video structure by using one or more spectrum variables.
8. A method of analyzing videos, said method comprising:
providing a digital database comprising a plurality of digital video files;
selecting one of said digital video files for analysis;
if necessary, transforming said one of said digital video files into a selected format comprising sequences of frames or photograms and audio files;
processing said one of said digital video files to extract video descriptors and metadata;
processing said one of said digital video files to establish correlations among said video descriptors;
establishing an output measure that relates said video descriptors with audience data, said establishing an output measure that relates said video descriptors with audience data further comprising:
tagging each scene with a time stamp to relate each scene to rating data;
evaluating said video descriptors for each scene at each time stamp; and
assigning a relative measure for each video descriptor correlated to said time stamp.
9. The method according to claim 8 , said evaluating said video descriptors for each scene further comprising:
processing said one of said digital video files to calculate statistics for parameters selected from the group consisting of:
amount of red,
amount of green,
amount of blue,
saturation,
light level/luminosity,
tone,
light level period, and
Hu's Momentum;
calculating tone and saturation histogram;
evaluating structures made up of lines to determine visual complexity of an image;
calculating an amount of movement between two consecutive frames/photograms;
analyzing scene frequency in grey scale, and for components in red, green, and blue;
measuring scene complexity using color segmentation; and
analyzing audio parameters selected from the group consisting of:
sound power,
spectrum fractiles from low frequencies,
maximum frequency and frequency centroid,
power of low, medium, and high frequencies,
power break, and
harmonic detector.
10. The method according to claim 8 , further comprising:
for each minute of each scene, correlating information selected from the group consisting of:
type and intensity of all scenes present in each minute,
characters that appear in each minute,
number of seconds of commercials, if any,
number of seconds of introductory material, if any,
minimum, maximum, mean and standard deviation of video descriptors of frames belonging to each minute, and
audience rating values for each minute;
establishing meaningful pairs of elements from said information; and
calculating a relative rating value for each video descriptor.
11. The method according to claim 10 , further comprising classifying said one of said digital video files in sets or clusters of videos having similar perceptual and cognitive, temporal parameters.
12. The method according to claim 11 , further comprising measuring deviation from a trend in one or more videos through detection of cycles that differ from said trend and detection of innovation and prediction wavelets.