Lai-Kuan Wong

Saliency-enhanced image aesthetics class prediction

We present a saliency-enhanced method for the classification of professional photos and snapshots. First, we extract the salient regions from an image by utilizing a visual saliency model. We assume that the salient regions contain the photo subject. Then, in addition to a set of discriminative global image features, we extract a set of salient features that characterize the subject and depict the subject-background relationship. Our high-level perceptual approach produces a promising 5-fold cross-validation (5-CV) classification accuracy of 78.8%, significantly higher than existing approaches that concentrate mainly on global features.

Enhancing visual dominance by semantics-preserving image recomposition

We present a semi-automatic photographic recomposition approach that employs a semantics-preserving warp of the input image to enhance the visual dominance of the main subjects. Our method uses the tearable image warping method to shift the subjects against the background (and vice versa), so that their visual dominance is improved, and yet preserve the desired spatial semantics between the subjects and the background. The recomposition is guided by a measure of the resulting visual dominance of the main subjects. Our user experiment shows the effectiveness of the approach.

Pressure-based touch positioning techniques for 3D objects

Most of the previous 3 DOF (Degree Of Freedom) 3D touch positioning techniques require more than one finger (usually two hands) to be performed, which limits their using space on small mobile devices such as phones and tablets that need one hand to be held in most occasions. Given that the pressure sensitive touch screen would become ubiquitous in near future, we present the pressure-based 3DOF 3D objects positioning and manipulating techniques that only use one hand in operating.

Aesthetic Evaluation of Facial Portraits Using Compositional Augmentation for Deep CNNs

Digital facial portrait photographs make up a massive portion of photos in the web. A number of methods for evaluating the aesthetics of photographs have been proposed recently. However, there have been a little work in the research community to address the aesthetics of targeted image domain, such as portraits. This paper introduces a new compositional-based augmentation scheme for aesthetic evaluation of portraits by well-known deep convolutional neural network (DCNN) models. We present a set of feature augmentation methods that take into account compositional photographic rules to ensure that the aesthetic in portraits are not hindered by standard transformations used for DCNN models. On a portrait subset of the large-scale AVA dataset, the proposed approach demonstrated a reasonable improvement in classification performance over the baseline and vanilla deep learning approaches.

The design and empirical evaluations of 3D positioning techniques for pressure-based touch control on mobile devices

The previous three degrees of freedom (DOF) 3D touch translations require more than one finger (usually two hands) to be performed, which limits their usability on mobile devices that need one hand to be held in most occasions. Given that the pressure-sensitive touch screen will become ubiquitous in the near future, we presented a pressure-based 3DOF 3D positioning technique that only uses one finger in operating. Our technique collects the normal force of the touch pressure and uses it to represent the depth value in 3D translating. Then we conducted several groups of tightly controlled user studies to conclude (1) how different strategies of pressure recognition will affect 3D translating and (2) how is the performance of the pressure-based manipulation compared to the previous two-fingered technique. Finally, we discussed some guidelines to help developers in the design of the pressure-sensing technique in 3D manipulations.

Towards Demographic-Based Photographic Aesthetics Prediction for Portraitures

Do women look at aesthetics differently from men? Does cultural background have an influence over the perception of beauty? Has age have any role in this? Psychological and art studies reveal striking differences in perception among various demographical aspects. This warrants attention particularly with the rapid growth in automatic evaluation of photo aesthetics. In this research, we investigate the influences of demographic factors of photographer towards the aesthetic quality of portrait photos from the computational perspective. An extended version of the large-scale AVA dataset was created with the inclusion of the photographers’ demographic data such as location, age and gender. We trained several demographic-centric CNN models, which are then fused together as a single multi-demographic CNN model to learn aesthetic tendencies in a holistic manner. We demonstrate the efficacy of our model in achieving state-of-the-art performance in predicting portraiture aesthetics.

Vehicle semantics extraction and retrieval for long-term carpark video surveillance

Car park video surveillance data provides plenty of semantic rich data such as vehicle color, trajectory, speed, and type which can be tapped into and extracted for video and data analytics. We present methods for extracting and retrieving color and motion semantics from long term carpark video surveillance. This is a challenging task in outdoor scenarios due to ever-changing illumination and weather conditions, while retrieval time also increases as data size grows. To address these challenges, we subdivided the search space into smaller chunks by introducing spatio-temporal cubes or atoms, which can store and retrieve these semantics at ease. The proposed method was tested on 2 days of continuous data from an outdoor carpark under various lighting and weather conditions. We report the precision, recall and \(F_1\) scores to determine the overall performance of the system.

Pic2Geom: A Fast Rendering Algorithm for Low-Poly Geometric Art

Low poly rendering has always been a popular form of art in the art and design community, where the artist draws each polygon (usually triangle) on the image individually. Several state-of-the-art methods were proposed to overcome the laborious process of manual low poly rendering, by rendering the low poly shapes automatically. However, results generated by the aforementioned methods were either not visually pleasing or the algorithms are computationally slow. In this paper, we present Pic2Geom, a fast algorithm that generates low poly geometric art with adequate quality, at low computational cost. The proposed algorithm utilizes edge detection, saliency detection and face detection to generate a set of seed points, which is then used by Delaunay triangulation to generate the low poly abstraction. Comparison with state-of-the-arts approaches demonstrate the efficiency and effectiveness of our algorithm in producing a quality geometric abstraction of a given photograph.

Automatic Detection and Counting of Circular and Rectangular Steel Bars

The steel industry heavily relies on manual labor and the use of photoelectric sensors and complex counting machines to count steel bars. In the last decade, research on the automatic detection and counting of steel bars by using image processing and computer vision techniques have seen much progress. Nevertheless, most of past research focused mainly on circular shaped steel bars from a direct frontal camera angle. In this paper, we propose a method that is adaptable to both circular and rectangular shaped steel bars, and robust towards different camera angles and lighting intensity. The captured digital image first undergoes an essential pre-processing stage followed by edge detection which extracts the steel bar edges. For circular shaped steel bars, we apply Hough Transform followed by a post-process while the rectangular ones can be accurately found based on a series of morphological operations. Experiments conducted on a variety of challenging conditions demonstrate the capability of our approach to a good measure of success.

Semantics-Preserving Warping for Stereoscopic Image Retargeting

Due to availability and popularity of stereoscopic displays in the recent years, research into stereo image retargeting is receiving considerable attention. In this paper, we extend the tearable image warping method for stereo image retargeting. Our method retargets both the left and right image of the stereo image pair simultaneously to preserve scene consistency, and minimize distortion using a global optimization algorithm. It is also able to preserve stereoscopic properties of the resulting stereo image. Experimental results show that our approach can preserve the global image context better than stereoscopic cropping, preserve structural details better than stereoscopic seam carving, and protect objects better than stereoscopic traditional warping. Besides, compared to scene warping, our approach can guarantee semantic connectedness.