Minimum average correlation energy filters

Mahalanobis, Abhijit; Kumar, B. V. K. Vijaya; Casasent, David P.

doi:10.1364/ao.26.003633

Cited by 612 publications

(293 citation statements)

References 9 publications

Supporting

Mentioning

290

Contrasting

Unclassified

Order By: Relevance

“…The algorithm of correlation filters [10] in its unconstrained form [11] is simply linear least-squares regression applied to the set of all circular shifts of every example. While including circular shifts in the training set may seem peculiar, it leads to a system of equations which can be constructed and solved in the Fourier domain.…”

Section: Correlation Filtersmentioning

confidence: 99%

“…Adopting this assumption within an LDA framework, comparable detection performance to HNM has been demonstrated [9]. The second method is that of correlation filters [10,11], in which all circular shifts of each example are incorporated into the training set (see Figure 2). This manifests in a covariance matrix which is not only Toeplitz, but circulant…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Learning Detectors Quickly with Stationary Statistics

Valmadre

Sridharan

Lucey

2015

Computer Vision – ACCV 2014

View full text Add to dashboard Cite

Abstract. Computer vision is increasingly becoming interested in the rapid estimation of object detectors. The canonical strategy of using Hard Negative Mining to train a Support Vector Machine is slow, since the large negative set must be traversed at least once per detector. Recent work has demonstrated that, with an assumption of signal stationarity, Linear Discriminant Analysis is able to learn comparable detectors without ever revisiting the negative set. Even with this insight, the time to learn a detector can still be on the order of minutes. Correlation filters, on the other hand, can produce a detector in under a second. However, this involves the unnatural assumption that the statistics are periodic, and requires the negative set to be re-sampled per detector size. These two methods differ chiefly in the structure which they impose on the covariance matrix of all examples. This paper is a comparative study which develops techniques (i) to assume periodic statistics without needing to revisit the negative set and (ii) to accelerate the estimation of detectors with aperiodic statistics. It is experimentally verified that periodicity is detrimental.

show abstract

Section: Correlation Filtersmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Learning Detectors Quickly with Stationary Statistics

Valmadre

Sridharan

Lucey

2015

Computer Vision – ACCV 2014

View full text Add to dashboard Cite

show abstract

“…There are many different types of correlation filters, and the differences lie in how the filters are formed. Examples of correlation filters include the Minimum Average Correlation Energy (MACE) filters [13], the Optimal Tradeoff Filters (OTF) [18], the Unconstrained Minimum Average Correlation Energy (UMACE) filter [20], the Average of Synthetic Exact Filters (ASEF) [3], the Synthetic Discriminant Functions (SFD) [11] , the Minimum Variance Synthetic Discriminant Functions (MVSDF) [12], and the Adaptive Average Correlation Energy (AACE) [21]. To perform this task we have opted to use an UMACE filter.…”

Section: A Facial Feature Detection Using Image Correlationmentioning

confidence: 99%

Correcting rolling-shutter distortion of CMOS sensors using facial feature detection

Heflin

Scheirer

Boult

2010

2010 Fourth IEEE International Conference on Biometrics: Theory, Applications and Systems (BTAS)

View full text Add to dashboard Cite

Abstract-This paper proposes a fully automated post image processing scheme based on facial feature detection to correct the horizontal temporal shear or rolling shutter distortion. This distortion occurs when obtaining images or video sequences from a CMOS camera with a rolling shutter whenever there is relative horizontal movement between the sensor and the object being imaged during the integration time of the image frame. Unlike CCD sensors, such as the interline CCD, which provides an electronic shutter mechanism called a global shutter in which the light collection starts and ends at exactly the same time for all pixels, CMOS sensors can not hold and store all the pixels at the same time. Each scanline is exposed, sampled, and stored in sequence, resulting in the rolling shutter effect or temporal distortion of the image that will cause inaccurate facial recognition results.Facial feature detection is performed using correlation based methods with low computational complexity. The location of key facial feature points is then used to calculate the temporal horizontal shear or the distortion of the image. This information can then be used to remove the temporal horizontal shear distortion from the detected face or the entire image. We present experimental results on controlled data sets and real scenes to show that the proposed method yields excellent results in reversing the temporal horizontal shear caused by the CMOS rolling shutter sensor and significantly improves the accuracy of our facial recognition algorithm.

show abstract

“…This makes it difficult to determine the location of shifted targets. Minimum average correlation energy (MACE) filters 80 were proposed to solve such a problem. Recently, a new algorithm based on the MACE concept that allows the selection of the spatial frequency content of the filter automatically, has been proposed.…”

Section: Composite Filtersmentioning

confidence: 99%