nils-holmberg/cca-cv

computational content analysis, computer vision

HTML

computational content analysis, computer vision

visuals as content (and some reception of visuals later on)
content analysis is under-utilized in communication research ?
human coders are the gold standard, but does not scale up

projects

automated web scraping

digital images, videos, web pages, source code, metadata
dont forget subtitles, automatic transcription of video

low-level content analysis

cf. "manifest content" in web pages, videos
videos need to be split into frames, images
pixels, color histogram, dimensions, edges

image feature extraction

cf. "latent content", infer content features
low, mid, high level features, lower are easier

image saliency is mid-level feature? more computationally intensive..

image similarity

understand, infer relationships between images
image embeddings, vector space, topic modelling for images

image classification

high-level content features, object categories inferred from pixels
does image contain faces, people or not, binary classification

try an image classifier
some instructions here
open question: when could binary image classification be useful in communication research ?
e.g. distinguish content from advertising..

object detection

detect multiple basic object classes, aka labelling
face identification is a special case of this approach
typically also involves object localization
get bounding boxes of multiple objects

try it at google vision ai
try it at ms computer vision
open question: when could object detection be useful in communication research ?

semantic segmentation

outline actual pixels covered by each object in image
web page segementation, (quite difficult!)

open question: when could object detection be useful in communication research ?
e.g. large scale video analysis..

pose estimation

infer body parts and relationships
if applied to face images, becomes image sentiment analysis

optical character recognition

extract text from images, screenshots, etc

image captioning

could be used as input to word vectors

open question: when could image captioning be useful in communication research ?

video summarization

actions, actors, shot detection, data reduction

google video ai

natural language processing, nlp

not in the visual domain, so not dealt with here

cognitive communication effects

reception of visuals
media psychology

survey experiments

advert content features and product attitude

try demo experiment

lab experiments

product visual saliency and visual attention (eye-tracking)

computational content generation

production of visuals

data visualization

facilitate understanding, comprehension of big data

generate new images

gans, artwork, people, dall-e 2 (not content analysis)

this person does not exist

augmented, virtual reality

avatars, virtual models