Content analysis of video using principal components

Abstract

We use principal component analysis (PCA) to reduce the dimensionality of features of video frames for the purpose of content description. This low-dimensional description makes practical the direct use of all the frames of a video sequence in later analysis. The PCA representation circumvents or eliminates several of the stumbling blocks in current analysis methods and makes new analyses feasible. We demonstrate this with two applications. The first accomplishes high-level scene description without shot detection and key-frame selection. The second uses the time sequences of motion data from every frame to classify sports sequences.

Classification of sport videos using edge-based features and autoassociative neural network models

“…The AANN based classifier has better generalizing ability than the k-nearest neighbor classifier. Similarly, a time-constrained clustering algorithm [13] using compressed color features requires a minimum of 50 s of test data to yield a classification performance comparable to the proposed method. The proposed method was applied on test clips of 20 s duration in all the experiments on video classification.…”

“…Apart from the duration of the test data, the quality of the test data also influences the classification performance. Some methods retain only the class-specific frames in the test data by editing out images related to crowd/audience or off-field action [13]. Such editing results in an improved performance.…”

“…Several techniques for video classification typically use test clips with durations varying from 60 to 180 s [6,5,9,13,14,31]. The classification performance was found to improve with increase in the duration of the test clip.…”

“…The probability density function of the compressed features is then modeled using a mixture of Gaussian densities. Dimension reduction of low-level features such as color and texture, using PCA, has also been attempted in [12,13]. Another approach, described in [14], constructs two hidden Markov models (HMMs), one from the principal motion direction, and the other from the principal color of each frame.…”

Detecting molecular separation in nano-fluidic channels through velocity analysis of temporal image sequences by multivariate curve resolution

“…The higher order components retain information on changes. For example, in IR image sequences higher PCs showed a dynamic scene behavior during time variant heating by solar illumination (Sahouria and Zakhor 1999). All two-way multivariate analysis methods such as PCA and MCR suffer from rotational ambiguities (Malinowski 1991;Tauler et al 1995).…”