Smart Scribbles for Sketch Segmentation
Summary (3 min read)
1. Introduction
- Sketchy drawings are prevalent across a wide range of applications and domains.
- Standing of the content of the drawing is required.
- To-date, there is no efficient method available for automatic segmentation in this domain.
- The authors seek a semi-automated solution to segmenting sketchy drawings that is fast enough for interactive use, but also predictable and easy to use – making it accessible to even the most novice user.
- In contrast to previous methods that use scribbles as positional constraints for various image editing tasks [BJ01,LLW04,AP08, SDC09b], their formulation considers more detailed geometric (position, orientation, curvature) and temporal information (time of creation) when analyzing stroke-to-stroke and Scribble-to-stroke relationships.
3. Method
- The method the authors present allows users to intuitively segment digital sketches into semantically meaningful regions.
- This helps us to differentiate strokes which are spatially close but are drawn at different moments in time.
- The input Scribbles are special strokes that indicate the user’s intent to segment a particular portion of the drawing.
- If desired, the user should be able to precisely select localized regions.
- The concept of their design is depicted in Fig. 2 and the remainder of this section describes in detail each of the steps used by their method.
3.1. Energy function
- The data term Di measures the affinity between Scribbles and strokes.
- The parameter λ controls the relative influence of the smoothness and data terms.
3.1.2. Data Term
- The affinity A(i,r) is defined as: A(i,r) = ∏ g∈Gdata δ(g(i,r),σg) (5) Here, as with the smoothness term, the authors measure the similarity between segments rather than strokes.
- To illustrate this, the authors consider a scenario where the user draws a single Scribble, as shown in Fig. 3a.
- This behavior, though reasonable, is not in line with a user’s expectations of having local control.
- This rule was used to control the selection locality in systems having limited modality [LS05].
- Since the spatial spread of the fall-off function (6) grows linearly with the increasing σprox the authors can set σprox = W/2.
3.2. Optimization method
- The graph vertices V consist of stroke segments S and label terminals L. Each stroke segment i ∈.
- In addition, auxiliary edges Ei,l connect stroke segments i ∈ S to label terminals l ∈ L. Each Ei,l has weight wi,l = λ(1−Di(l)), where λ is the parameter defined in Equation 1. c© 2012 The Author(s) c© 2012 The Eurographics Association and Blackwell Publishing Ltd.
- The multi-way cut problem with two terminals is equivalent to a max-flow/min-cut problem for which efficient polynomial algorithms exist [BK04].
- For three or more terminals the problem is NP-hard [DJP∗92].
- This approach provides results similar to more advanced techniques (such as α-expansion or α/β-swap [BVZ01]), but is significantly faster and therefore better suited for interactive applications.
4. Results
- The authors demonstrate the effectiveness of their algorithm on a variety of input sketches.
- These results show that desirable sketch segmentations can be obtained using very different scribbling strategies.
- The authors framework does not require artists to draw the input sketches in any particular manner.
- It is possible that strokes representing the same region can be drawn at very different moments in time.
- This can diminish the advantage of taking timing into account in the similarity metric.
4.1. User Study
- Paired t-tests (see Table 2) indicate that this gain is statistically significant (p < 0.005) for 4 out of 6 drawings.
- Although the median speed-up is apparent, the advantage of Smart Scribbles is not as convincing in this case.
- The main reason here could be the relatively low complexity of these examples.
- In addition, participants were presented with different ways of controlling the locality.
5. Applications
- The labeling produced by their approach can be utilized to generate input to perform region as well as stroke segmentation (see Fig. 10a).
- This image is used both as an input gray-scale image (Fig. 9a) and as foreground soft scribbles (blue in Fig. 9b) for input to LazyBrush.
- Given this input, LazyBrush produces the desired area mask (Fig. 9c).
- The ability to easily label both strokes and areas empowers a large variety of applications.
- If available, the authors can use these aiding structures as Scribbles to segment the final detailed sketch (see Fig. 10c).
7. Conclusions
- The authors have presented Smart Scribbles, a scribble-based interface for sketch segmentation.
- The authors method is fast, supports multi-label segmentation, and acts as an enabling technology for a variety of applications in the context of drawing, editing, and animation.
- In the long term, the authors envision a next-generation drawing application, where drawing, editing, and animation are tightly integrated, and where the simplicity of the interaction is the key.
- This work represents a step in this direction; a bridge between classic drawing and digital editing.
Did you find this useful? Give us your feedback
Citations
200 citations
Cites methods from "Smart Scribbles for Sketch Segmenta..."
..., constructed in layers) or can be easily obtained with interactive tools [Noris et al. 2012]....
[...]
...Such segmentations may come with the drawing itself (e.g., constructed in layers) or can be easily obtained with interactive tools [Noris et al. 2012]....
[...]
90 citations
88 citations
Cites background or methods from "Smart Scribbles for Sketch Segmenta..."
...(Refer to [Noris et al. 2012] for more algorithm details....
[...]
...for experienced artists and thus defeats the main motivation behind sketch-based interfaces [Noris et al. 2012]....
[...]
...(Refer to [Noris et al. 2012] for more algorithm details.)...
[...]
...We use graph cut to optimize the labeling, similar to [Noris et al. 2012]....
[...]
...…components one by one in a specific order), which however is often an error-prone process even ACM Transactions on Graphics, Vol. 33, No. 6, Article 175, Publication Date: November 2014 for experienced artists and thus defeats the main motivation behind sketch-based interfaces [Noris et al. 2012]....
[...]
64 citations
58 citations
Cites background from "Smart Scribbles for Sketch Segmenta..."
...A semi-automatic approach could be envisioned here were the user would scribble ([Noris et al. 2013]) or sketch small primitives to help both resolve the correspondence and specify the choice of internal body line....
[...]
References
7,413 citations
4,463 citations
3,571 citations
3,252 citations
3,199 citations
Related Papers (5)
Frequently Asked Questions (17)
Q2. What have the authors stated for future works in "Smart scribbles for sketch segmentation" ?
In the future the authors plan to develop a system that allows automatic parameter tuning based on a database of ground truth data. The use of previously labeled drawings as Scribbles offers another avenue for future work. However, in the worst case, when a large number of stokes are close to each other as defined by their similarity measure, the number of edges in the graph can grow quadratically with the number of strokes and the computation can become prohibitively slow ( see Fig. 13 ). The problem can be alleviated by subsampling the strokes and processing disconnected components individually.
Q3. What is the main advantage of their work?
the main advantage of their work is that the authors formu-c© 2012 The Author(s) c© 2012 The Eurographics Association and Blackwell Publishing Ltd.late an energy minimization problem where, in addition to shape similarity, the authors also take proximity, orientation, temporal information, and smoothness of the final labeling into account.
Q4. What are the main categories of prior art?
Relevant prior art can be divided into three main categories: sketch labeling interfaces, scribble-based image segmentation, and classification of vector fields in scientific visualization.
Q5. What are the limitations of the automatic technique?
Two limitations of this automatic technique are 1) limited complexity of objects detected by the system and 2) the inability to handle sketchy overlapping curve fragments, thus requiring more traditional and tedious lasso/selection-box methods for more complex drawings.
Q6. How is S connected to other stroke segments?
S is connected to all other stroke segments j ∈ S−{i} via edges Ei, j having weight wi, j equal to the smoothness term Vi, j when φi 6= φ j.
Q7. What is the main idea of the paper?
The approach presented in this paper leverages previous works on interactive image segmentation in order to optimize the labeling process based on user scribbles.
Q8. What is the common problem with the current approach?
Scribbles could also potentially be used to improve the accuracy of drawing simplification methods [GDS04,BTS05,SC08], as a typical problem with the current, fully automatic, approaches is that they do not take into account any semantic information such as provided by their approach.
Q9. How do the authors control the locality of each Scribble?
the authors control the locality of each Scribble r by modifying its proximity fall-off δ (3) as follows:δ(prox(i,r),σprox) = 1σprox exp( − prox(i,r) 2σ2prox) .
Q10. Why does the framework remove clutter from the input drawing?
Because this process removes clutter from the input drawing, it greatly improves the accuracy of selection and reduces the amount of user interaction needed to obtain clean results.
Q11. What is the purpose of the sketch clustering?
The remainder of this section describes in detail each of the steps used by their methodThe authors formulate the task of sketch clustering as an optimization problem, where the goal is to label each stroke in a way that minimizes an energy function.
Q12. What is the way to reduce the number of strokes in the graph?
in the worst case, when a large number of stokes are close to each other as defined by their similarity measure, the number of edges in the graph can grow quadratically with the number of strokes and the computation can become prohibitively slow (see Fig. 13).
Q13. What is the input to the minimization method used to compute the final solution to the stroke label?
The input to their method consists of a set of stroke segments S and a set of Scribble segments R associated with a set of labels L.
Q14. What is the simplest heuristic for the multi-way cut problem?
To obtain a good approximate solution the authors use a simple divideand-conquer heuristic previously proposed in [SDC09b] to gradually simplify the N-terminal problem into a sequence of N − 1 binary max-flow/min-cut sub-problems.
Q15. What is the way to reduce the number of augmentation paths?
In certain situations where the cost of the minimal cut is very high and the graph topology is complex, the number of augmentation paths can grow very quickly along with the computation time.
Q16. What is the equivalent of a max-flow/min-cut problem?
The multi-way cut problem with two terminals is equivalent to a max-flow/min-cut problem for which efficient polynomial algorithms exist [BK04].
Q17. How can the authors use Scribbles to improve the efficiency of drawing?
This approach could be used, for instance, as an extension to the recently presented ShadowDraw system [LZC11], by augmenting each sketch in the database with Scribbles.