Youngjun Yoo

Color filter array interpolation using color correlations and directional derivatives

We present a color filter array interpolation algorithm to provide the missing colors associated with the commonly used Bayer pattern in digital still camera systems. The developed algo- rithm utilizes color correlation and directional derivative attributes of neighboring color pixels and is thus named C2D2. The main advan- tage of this algorithm is that it requires no user-defined parameter, as is the case in many edge-adaptive color filter array interpolation algorithms. Both an objective and subjective comparison are carried out with two other popular edge-adaptive algorithms. The results indicate that this algorithm consistently generates lower mean- squared interpolation errors in the RGB and L*a*b* color spaces.

Programmable DSP platform for digital still cameras

This paper presents a programmable DSP platform for digital still cameras based on the Texas Instruments TMS320C54x family. One major advantage of this platform is that, after capturing an image from a CCD sensor, processing the raw image, and compressing the image for storage is performed on the digital signal processor (DSP). This provides a short shot-to-shot delay and a high degree of flexibility. The system realized also allows instant viewing and selective storing of captured images. This paper outlines the various processing stages necessary to take the raw CCD data and produce a JPEG compressed bit stream and highlighting the advantages of DSPs for this application. The programmable nature of this platform allows for the exploration of different image processing and compression techniques. The low power nature of the digital signal processor provides long battery life.

Adaptive quantization of image subbands with efficient overhead rate selection

Subband image coding techniques owe much of their success to an effective use of adaptive quantization and adaptive entropy coding. It is often the case that adaptive quantization is achieved by defining a discrete set of quantizers from which one is chosen for a given set of coefficients. This type of forward adaptation thus requires that overhead information (the choice of quantizer) be sent to the decoder. Then, the quantized coefficients are transmitted using adaptive entropy coding, typically through backward adaptive arithmetic coding. We show that a combination of forward and backward adaptation methods can be used to update the quantizers thus reducing the overhead requirements while still providing good performance. Specifically, we present an algorithm where each coefficient is classified into several classes based on the past quantized data and where the quantizer to be used for each class can itself be adapted on the fly.

Development and real-time implementation of auto white balancing scoring algorithm

This paper presents the development and real-time implementation of an autowhite balancing algorithm named scoring. The spectral distributions of the Macbeth reference colors together with the spectral distributions of several color temperature light sources are used to set up a number of reference color points in the CbCr color space. A number of representative color points are obtained from a captured image by using a previously developed multi-scale clustering algorithm. A match is then established between the reference set of colors and the representative set of colors. The matching scheme generates the most likely color temperature under which the image is captured. Furthermore, this paper discusses the real-time implementation of the developed auto white balancing algorithm on the TI TMS320DSC platform, a power-efficient single-chip processor that has been specifically designed for digital still cameras. It is shown as to how the algorithm is modified to allow a processing rate of 30 frames/s.

Real-time implementation of auto white balancing and auto exposure on the TI DSC platform

This paper presents the real-time implementation of an auto- white-balancing and an auto-exposure algorithm on the TI TMS320DSC21 platform. This platform is a power-efficient single-chip processor that has been specifically designed for digital still cameras. Its architecture consists of five subsystems including an ARM micro-controller, a DSP core, a memory subsystem, two co-processors, and an imaging peripherals subsystem. Due to the memory constraints, the algorithms were modified to allow their real-time implementation of the processor, i.e. a processing rate of 30 frames per second. These modifications are discussed in the paper. The details of the algorithms are reported in an accompanying paper presented in the SPIE conference 4669B.

New approach to auto-white-balancing and auto-exposure for digital still cameras

This paper presents an auto-white balancing algorithm named scoring. The spectral distributions of the Macbeth reference colors together with the spectral distributions of various color temperature light sources are used to obtain a number of reference color points in the CbCr color space. A number of representative color points are also obtained from a captured image by using a previously developed multi-scale clustering algorithm. A match is then established between the set of reference colors and the set of representative colors. The matching scheme generates the most likely light source candidate under which the image is taken. Furthermore, this paper presents an auto-exposure algorithm using a mapping from the luminance histograms of five sub- areas in the image to an exposure value. A neural network is designed to perform the mapping. The histogram in each sub- area is used to determine the mean, variance, minimum, and maximum luminance for that sub-area. The same spatial information is computed for previous frames to incorporate temporal changes in luminance into the network.

Constrained bit allocation for error resilient JPEG coding

Variable-rate bit streams generated under the current DCT-based JPEG standard are vulnerable to propagation of error across the data units, i.e., DCT blocks. A traditional approach to combat this problem has been to introduce resynchronization markers in the transmitted bitstream. We propose an alternative, and more efficient, approach based on constraining the data length for the blocks or groups of blocks in the compressed image. In the proposed scheme the number of coded bits in each block, or group of blocks, is sent to the decoder as side information and thus the decoder can introduce resynchronization markers. The required overhead is significantly less than that of typical explicit synchronization schemes (e.g., those in JPEG). In addition, it is possible to further constrain the allowable rates for the blocks in order to reduce the overhead. We show experimental results where the constrained rate allocation is performed using thresholding techniques on images encoded by the baseline JPEG.

Adaptive quantization without side information using scalar-vector quantization and trellis coded quantization

We combine backward adaptive quantization with the scalar-vector quantizer (SVQ) and the trellis coded quantizer (TCQ) both of which have an underlying scalar quantizer (USQ) in their structure. The resulting adaptive scalar-vector quantizer (ASVQ) and adaptive trellis coded quantizer (ATCQ) redesign the USQ based on the past quantized outputs. The adaptive quantizers require no side information and also outperform the SVQ and the TCQ, respectively, when the input signal is non-stationary. For an input sequence from a bimodal source switching infrequently between two Gaussian distributions with the same mean and different variances, both adaptive quantizers achieve performance gains of more than 1.3 dB over the non-adaptive quantizers designed on the training set from the same bimodal source. Also the adaptive quantizers demonstrate minimal performance degradation due to adaptation when stationary inputs are considered.