Ravi Shenoy

Subspace based disparity estimation for plenoptic cameras

We present a subspace based disparity estimation technique for plenoptic 2.0 lightfield cameras. The raw lightfield image contains a micro-image for every lens in the micro-lens array. The disparity of a scene point is typically estimated using multi-baseline approach. The multi-baseline approach necessitates that a focussed copy of a patch is present in at least one of the neighboring micro-images. This requirement limits the range over which the disparity can be reliably estimated. We propose a subspace based technique for disparity estimation wherein a subspace for every disparity is learnt separately, and the learnt subspaces are subsequently used for estimating the disparity of any micro-image. We estimate the disparities for the images captured using Raytrix R11 camera and compare the results with (a) estimates obtained from the multi-baseline approach, and (b) manufacturer provided disparity maps. Comparisons show that the disparity maps estimated by the proposed technique are superior. In addition, the proposed technique allows for extending the range over which the disparity can be estimated.

Frequency domain linear prediction based on temporal analysis

Frequency-domain linear prediction (FDLP) is widely used in speech coding for modeling envelopes of transients signals, such as voiced and unvoiced stops, plosives, etc. FDLP fits an auto regressive model to the discrete cosine transform (DCT) coefficients of a sequence. The spectral prediction coefficients provide a parametric model of the temporal envelope. The prediction coefficients are obtained by solving the set of Yule-Walker equations expressing the relationship between lagged spectral autocorrelation values. A limitation of the direct approach of computing the spectral autocorrelation values is that the sequence has to be padded with a large number of zeros for the autocorrelation estimates to be reasonably accurate. This comes at the cost of increased computational complexity. We present an efficient and accurate method for computing the spectral autocorrelation samples. We show that the spectral autocorrelation can be computed as cosine-weighted temporal centroids, where the weighting function is dependent on time-index of the samples.