Image Segmentation Using a Sparse Coding Model of Cortical Area V1

Abstract

Algorithms that encode images using a sparse set of basis functions have previously been shown to explain aspects of the physiology of primary visual cortex (V1), and have been used for applications such as image compression, restoration, and classification. Here, a sparse coding algorithm, that has previously been used to account of the response properties of orientation tuned cells in primary visual cortex, is applied to the task of perceptually salient boundary detection. The proposed algorithm is currently limited to using only intensity information at a single scale. However, it is shown to out-perform the current state-of-the-art image segmentation method (Pb) when this method is also restricted to using the same information.

A simplified texture gradient method for improved image segmentation

“…The performance obtained by integrating the proposed method with the PC/BC model of V1 (F-score 0.64) is slightly better than the performance obtained by the Pb algorithm (F-score 0.63) [5]. The reason that the combined algorithm, outperforms the Pb algorithm is that the PC/BC model of V1 produces better performance in detecting boundaries defined by local intensity discontinuities (F-score 0.61) [10] than the brightness gradient method used in the Pb algorithm (F-score 0.60) [5]. The current state-of-art algorithms for image segmentation are SCG [13] and gPb_ucm [14].…”

“…Despite the proposed simplifications, the performance of the proposed method in detecting texture boundaries is comparable with the texture gradient method used in the Pb algorithm when tested with the BSDS300 benchmark [9]. Furthermore, a combination of the proposed texture gradient method with the PC/BC model of V1 [10] enables boundaries defined by local discontinuities in both intensity and texture to be located. When tested with the BSDS300 benchmark, the combined algorithm slightly outperforms the current state-of-art image segmentation method (the Pb algorithm) when this method is also restricted to using only local discontinuities in intensity and texture at a single-scale within gray-scale images.…”

“…The following three modifications are required to integrate the boundaries derived from the proposed model with the boundaries derived from the PC/BC model of V1 [10].…”

“…Third, in step 5, after the maximum texture gradient is found at each pixel across the eight orientations, the maximum texture gradient is added to the boundaries derived from the PC/BC model of V1 [10] to yield a combined map. Afterwards, non-maximum suppression [11] is applied to thin the boundaries in this combined map.…”

“…Because the proposed method for locating boundaries is based on local discontinuities in texture alone, it cannot detect boundaries defined by intensity discontinuities. Spratling [10] has developed the predictive coding/bias competition (PC/BC) model of V1 that can be used to locate boundaries defined by intensity discontinuities within an image. It has been shown that the PC/BC model of V1 outperforms the Probability of boundary algorithm [5] when this algorithm is also restricted to using local discontinuities in intensity (brightness) at singlescale in gray-scale images.…”