Co-Sparse Textural Similarity for Interactive Segmentation

Abstract

Abstract. We propose an algorithm for segmenting natural images based on texture and color information, which leverages the co-sparse analysis model for image segmentation. As a key ingredient of this method, we introduce a novel textural similarity measure, which builds upon the cosparse representation of image patches. We propose a statistical MAP inference approach to merge textural similarity with information about color and location. Combined with recently developed convex multilabel optimization methods this leads to an efficient algorithm for interactive segmentation, which is easily parallelized on graphics hardware. The provided approach outperforms state-of-the-art interactive segmentation methods on the Graz Benchmark.

Interactive Full Image Segmentation by Considering All Regions Jointly

“…Several older works on interactive segmentation handle multiple labels in a single image [39,40,50,53]. We present the first interactive deep learning framework which does this.…”

Interactive volumetric segmentation for textile micro‐tomography data using wavelets and nonlocal means

“…When these properties are not enough to capture different phases of the object, considerable effort is required to design other priors. Examples include adding spatial dependency of the labeling using the distance transform , star‐convexity of the shape given the initial position of a labeling , texture features , and in other cases a combination of spatial cues, color, and dictionary learning .…”

“…Overall, distance cues combined with texture can improve segmentation. The model being that, if an unknown voxel is close to a given labeled voxel, and they present similar texture, they will also be more likely to share the same label [29]. This approach is often adequate if one needs to segment a single image and a single object, but loses its effectiveness when generalizing to unseen images with multiple instances of the same object.…”

“…All preprocessing steps are held constant. The features were chosen as our wavelet features are known to have a strong response to orientation, and LBP and gray-level features were previously reported to give the largest classification boost in natural images aside from a learned dictionary [29,40].…”

A weakly supervised geodesic level set framework for interactive image segmentation

“…Wu et al [43] formulate the interactive segmentation problem as a multiple instance learning (MIL) task by generating positive bags from pixels of sweeping lines within a user-provided bounding box. Nieuwenhuis et al [26] propose an algorithm based on texture and color information, which leverages the co-sparse analysis model for image segmentation within a convex multilabel optimization framework. CRF based algorithms to tolerate user input errors in scribble-based interactive segmentation have been proposed in [38,4].…”